AU AARHUS. Soil Environmental DNA as a Tool to measure Floral Compositional Variation and Diversity. Anne Aagaard Lauridsen

Transkript

1 Soil Environmental DNA as a Tool to measure Floral Compositional Variation and Diversity across Large Environmental Gradients Anne Aagaard Lauridsen June 2016 AU AARHUS UNIVERSITY Centre for GeoGenetics Natural History Museum of Denmark

2 Data sheet: Project: Scope: English title: Danish title: Master thesis in Biology 60 ECTS Soil Environmental DNA as a Tool to measure Floral Compositional Variation and Diversity across Large Environmental Gradients Jord-DNA som et værktøj til at måle floral kompositionel variation og diversitet over store miljømæssige gradienter Author: Anne Aagaard Lauridsen, Institution: Aarhus University Supervisors: Rasmus Ejrnæs Tobias Frøslev Delivered: 3 rd of June 2016 Defended: 21 st of June 2016 Key-words: Referencing style: Front page: edna, Soil edna, trnl, trnl p6 loop, Biodiversity, Community composition As in Molecular Ecology Pictures taken by Anne Aagaard Lauridsen and Jesper Stern Nielsen. Layout: Jesper Stern Nielsen Number of pages: 76 2

3 Preface This master thesis is the result of 60 ECTS point or one year s work. The aim of the study was to look at two different survey methods for plants, the soil environmental DNA sampling and the regularly used above ground survey, and whether they yielded similar conclusions regarding alpha diversity, community and taxonomic composition. The project was done under the wings of the large Biowide project, and was supervised by Rasmus Ejrnæs, Aarhus University, Department of Bioscience Biodiversity and Conservation, Grenåvej 14, 8416 Rønde, and Tobias Frøslev, Copenhagen University, Centre for GeoGenetics, Universitetsparken 15, 2100 København Ø. The thesis contains two parts: the main thesis manuscript written in a paper structure, and a post script, where work not included in the main manuscript has been briefly described. Abstract The purpose of this study was to test whether data produced from floral soil edna, sampled across large environmental gradients and differing soil types, could reliably reproduce lists of taxa, species richness and species composition found in a normal above ground (abg) survey of plant species. Plant edna using the p6 loop of the trnl region reproduced similar compositional patterns as a regular abg survey, and were able to detect the three natural gradients over which the sample sites were placed; nutrient level, successional stage and moisture level. The alpha diversity measured as OTU count from the edna trnl data correlated significantly with the species richness from the abg survey, but with very little explanatory power, thus not reliably reproducing known alpha diversity. The taxon retrieval varied, but up to 35.7% of families known to be present at the sites were retrieved, while less was retrieved for the genus and species level. Reference database issues, sampling design, marker choice, lab protocols and data processing needs to be revised in order to produce reliable biodiversity estimates. Resumé Dette studies formål var at teste om plante-edna (miljødna) fra jord, indsamlet langs store miljøgradienter og forskelligartede jordtyper, troværdigt kan finde eller estimere arts- eller taxonomi-lister, artsrigdom og sammensætning af de floristiske samfund på samme vis som en almindelig florainventering. Plante edna isoleret med trnl p6 loop-markøren reproducerede de floristiske samfundssammensætninger på samme måde som den almindelig flora inventering, og kunne samtidig detektere de underliggende miljøgradienter: næringsniveau, succcessionsstadie og fugtighedsniveau. Alphadiversitet målt ved OTU-antal fra trnl p6 loop edna data var signifikant korreleret med artsrigdom i den almindelige florainventering, men med meget lav forklaringsgrad, og kunne derfor ikke pålideligt estimere artsrigdommen i felterne. Taxongenkendelsen varierede, men edna-data genfandt op til 35,7 % af de familier man vidste var til stede i feltet. På arts- og genus-niveau var genfindelsesprocenten endnu lavere. Udfordringer med referencedatabaser, prøveindsamlingsdesign, markørvalg, laboratorieprotekoller og dataprocessering skal revideres for at kunne producere troværdige biodiversitetsestimater fra jord-edna. 3

4 Table of Contents Data sheet:... 2 Preface... 3 Abstract... 3 Resumé... 3 Soil environmental DNA as a tool to measure floral compositional variation and diversity across large environmental gradients... 6 Introduction... 6 Dictionary:... 6 Methods: Collection of samples in the field Sample sites Aboveground species list and environmental parameters Soil samples for edna DNA extraction, PCR and sequencing Database construction Raw data processing From raw data to OTU and sites by species table Above ground survey species lists and environmental variables Data analysis Diversity estimates Recovered taxa Results Ordinations Taxonomic assignment Diversity estimates The influence of physical and chemical parameters Discussion ID rate Problems with the reference data base False negatives and pipeline discrepancies Diversity measurements

5 Community composition Marker choice Conclusive remarks and future studies References Supplementary materials Taxonomic assignation and identification rates Rarefaction and species richness Ordinations Environmental fittings to ordinations Species lists Correlation plots between OTU count and environmental parameters Residuals and environmental parameters Scripts Script for reference database construction Script for raw data processing the OBITools pipeline Lab work, field work and other endeavors not included in the thesis manuscript Initial lab work Field work and subsequent lab work How to process vast amounts of data and learning three programming languages Time constraints References:

6 Soil environmental DNA as a tool to measure floral compositional variation and diversity across large environmental gradients Introduction Climate change, heavy land use, spread of invasive species and loss of habitat are affecting biodiversity worldwide, affecting ecosystem functionality and the ecosystem services they provide (Sala et al. 2009). The ability to monitor ecosystems and biodiversity on a large scale and at low cost is therefore much needed, both to observe where changes are most pronounced, and as a base for management decisions (McMahon et al. 2011; Yu et al. 2012). One way to monitor biodiversity at large scales is to focus on measurements of so-called biodiversity indicators, such as the IUCN red list index, developed to guide management to whether a species or group of species are threatened by extinction (Rodrigues et al. 2006; Butchart et al. 2011). More indicators can be measured by remote sensing, assessing species and structural diversity, forest tree species diversity as well as structural and habitat diversity (Dees et al. 2012; Leutner et al. 2012). However, the validity of using indicators to measure diversity at local and global scales has been an element of discussion, due to several factors not clearly understood such as the degree of intercorrelation between indicators, the representativeness of indices, whether indicators can be used to monitor species, the role of naturally rare species in such lists, and wrongly aimed management decisions, to mention a few (Quayle & Ramsay 2005; Butchart et al. 2006; Wiens et al. 2009; Nicholson et al. 2012; Yu et al. 2012). Traditional methods for inventorying biodiversity represent direct measurements but is time constrained by presence or activity of species (Hiiesalu et al. 2012) as well as time consuming and demands high taxonomic expertise (Chariton 2012).Thus there is a demand for rapid and cheap direct measurements of biodiversity, as may be accomplished through metabarcoding of the DNA found in environmental substrates (Taberlet et al. 2012b; Yoccoz et al. 2012; Ji et al. 2013). Dictionary: Fasta-file/multifasta-file: A file containing one or multiple sequences in a fasta format: A header line with sequence name or other information, followed by a new line containing the sequence. NGSfilter: A file containing information about which sample specific primer tags belongs to. It is used to assign the correct sample to the sequences from a sequencing run. Barcoding/metabarcoding: The term barcoding covers the process of isolating DNA from a single specimen, amplifying a taxa specific marker in order to obtain the species name or other taxonomic information of the specimen. Metabarcoding is essentially the same process, but performed on complex samples containing a mixture of species. Ideally, the idea is to use one single region in the DNA, and be able to distinguish to species level across the tree of life. In reality, markers specific for taxonomic groups are used, eg. plants, fungi or arthropods, and in general species level identification remains difficult at a global scale. OTU: Operational taxonomic unit, a group of sequence reads with identical nucleotide sequences. 6

7 Environmental DNA (edna) is a term covering DNA in various stages of degradation found in an environmental substrate such as soil, water or air, and can be used to identify the taxonomic level of organisms in the environment (Taberlet et al. 2012a). Such DNA consists of extracellular DNA, originating from naturally dead cells lost from living organisms, as well as intracellular DNA contained in still living cells (Levy-Booth et al. 2007). The term Barcoding, covering the use of isolated standard DNA markers to identify individual organisms, does not sufficiently describe the process of using edna for species identification in an environmental sample, since it contains DNA from multiple organisms and taxonomic groups (Taberlet et al. 2012a). To deal with the high taxonomic complexity of an environmental sample or mixtures of species from traps etc., the term DNA metabarcoding has been implemented and is widely used (eg. Epp et al. 2012; Yu et al. 2012; Ji et al. 2013). Environmental DNA was first addressed by microbiologists and has been used to examine microbial communities, taxa and functional traits for more than a decade (Giovannoni et al. 1990; Rondon et al. 2000), but edna as a tool in larger ecological studies really caught wind when Next Generation Sequencing (NGS) gained ground (Taberlet et al. 2012a). DNA from environmental samples has received more attention regarding the use in biodiversity monitoring in recent years, but started out as a method for examining the presence of extinct or extant species of birds, plants and mammals in ancient sediments (Willerslev et al. 2003). One of the first steps toward using edna as presence/absence indicators of modern biodiversity was taken by Ficetola et al. (2008) who detected the frog (Cana catesbeiana) in wetlands where it was known to be present. edna for biodiversity assessments has been used successfully in the aquatic environment (Lodge et al. 2012; Thomsen et al. 2012a; b), but Thomsen et al. (2012b) took a major leap forward, showing that edna from water samples can accurately measure biodiversity of fish, mammals, amphibians, insects and crustaceans present within the preceding two weeks. edna from soil samples has also stirred interest in biodiversity measurements as fungal edna diversity in forests were investigated by Buée et al. (2009), and earthworm diversity assessed by edna was shown to reflect diversity in a similar way as handsorting methods (Bienert et al. 2012). Andersen et al. (2012) were able to reconstruct overall known vertebrate taxonomic richness from soils collected in zoological gardens, and Yoccoz et al. (2012) found that soil edna represented ground survey plant taxonomic diversity well, both in boreal, temperate and tropical climates. Furthermore, Drummond et al. (2015) found that especially COI and 18S sequences from soil edna collected along an elevation gradient correlated well with regular measurements of aboveground biodiversity for trees, seedlings and invertebrates. Thus, soil edna can reflect both above- and below ground diversity, and may be very suitable for diversity monitoring in the future. DNA persistence in soil is influenced by biological, physical and chemical processes, determining the time frame a given DNA molecule can be retrieved from the soil (Levy-Booth et al. 2007). In a study on the persistence of crop-dna in soil from the French alps, crop DNA was found as long as 50 years back, but at low frequencies (Yoccoz et al. 2012). A German study on transgenic Beta vulgaris DNA, found that transgenic DNA could be detected up to two years after cultivation (Gebhard & Smalla 1999). In natural temperate environments it is likely that DNA will persist at least one year, depending on soil ph and composition, with acid soils adsorbing DNA better than alkaline soils (Levy-Booth et al. 2007). Since DNA accumulates in the soil over a period of time, edna analyses may be useful in identifying the yearly flora despite the shifting seasons, prolonging the fieldwork season into the fall and mild winters, where above ground surveys would prove difficult. Metabarcoding of soil may also be used to detect species that are difficult or undetectable by ground survey, such as dormant plants (Hiiesalu et al. 2012), nematodes 7

8 (Porazinska et al. 2009), collembola and other soil animals (Wu et al. 2009), fungi (Buée et al. 2009) or even early presence of invasive species, and simultaneously avoid the need for taxonomic identification expertise and observer bias (Yoccoz 2012). Soil edna may therefore provide an integrated picture of the diversity present at a given site. For many years, Sanger sequencing (F Sanger & AR Coulson 1975) has been the most precise, fastest and preferred DNA sequencing technology, but lately NGS technology has turned the table, allowing for large scale sequencing experiments including edna and metabarcoding studies. A Sanger sequencing reaction has the limitation of producing only one read, as opposed to NGS technologies producing millions of sequence reads in one reaction. Thus, metabarcoding and edna studies have accelerated after the invention of NGS (Taberlet et al. 2012a), allowing for the processing of multiple samples with multiple species in one reaction. As an example of the amounts of data being generated in NGS, the Illumina MiSeq sequencer can produce up to 15 million reads in one reaction (Illumina, 2016), posing new challenges with respect to data analysis. A broad bioinformatical workflow to analyze the millions of sequence reads from an environmental sample could be constructed as follows: errors from both amplification and sequencing are attempted removed, sequences are assigned to the sample of origin, identical sequences are gathered to operational taxonomic units (OTUs) and assigned to a taxonomic level before the creation of a sites by species table for further ecological interpretation (Coissac et al. 2012; Thomsen & Willerslev 2015). Several steps can be added or modified in this general pipeline, according to the nature of the target DNA region and question in mind (Coissac et al. 2012). One of the challenges in metabarcoding studies is the taxonomic assignation of the sequences, since the reference databases available for most ecosystems and organism groups are incomplete (Kvist 2013). Furthermore, for some markers, more species may share the same barcode, thus rendering species level identification impossible when only using a single barcode (Taberlet et al. 2007). Other reference database problems include possible naming confusion due to changing phylogenies and possible morphological misidentification of the species with barcodes present in the databases (Taberlet et al. 2007). A schematic overview of the workflow from environmental sample to species identification is vizualized in figure 1, where the bioinformatical pipeline is symbolized by the red arrow. Despite these to some extend unresolved problems, metabarcoding of environmental substrates have proven useful as a tool for measuring biodiversity. 8

9 Figure 1: Drawing of the workflow from environmental sample via DNA metabarcoding to species identification. The arrows indicate: Blue) different environmental substrates; ancient edna samples from glaciers, permafrost/tundra, aquatic sediments and modern samples from freshwater, terrestrial habitats and marine waters. Green) DNA extraction from the environmental samples. Orange) PCR amplification of the DNA extract using primers for targeting the desired taxa and sequencing of the amplified DNA on a NGS platform. Red) Bioinformatic processing of the data with error removal, sequence filtering and sorting, assigning taxa to sequences or OTUs and subsequent ecological interpretation of the resulting species/otus. Drawing from Thomsen & Willerslev (2015) by Lars Holm. When aiming to measure biodiversity and environmental changes, it is important to know whether edna can accurately recover a) presence of species, b) community composition, and c) alpha diversity. Despite promising outcomes, the road to using soil edna for biodiversity measurements is paved with difficulties regarding degradation, metabarcode length and specificity, as well as database and raw data processing issues. When analyzing ancient sediments as well as modern soils, the use of longer barcodes is difficult, as they are likely to result in less positive identifications due to the degraded state of DNA (Taberlet et al. 2007; Yoccoz et al. 2012; Little 2014). The barcodes suggested for land plants by the CBOL Plant Working Group et al. (2009), rbcl and matk ranges between bp (1000+ for some sequences of the rbcl marker), a marker length that is not likely to amplify well from degraded soils (Taberlet et al. 2007; Hollingsworth et al. 2011; Little 2014). In fact, a sum up from a range of papers on degradation and amplification success, Little (2014) concluded that the median recoverable plant DNA fragment is 190 bp long (IQR= bp), with an inverse relationship between amplification success and metabarcode length. However, shorter markers typically have a lower discriminatory power compared to longer markers, and 9

10 may yield a list of genera, families or orders instead of species (Willerslev et al. 2003; Taberlet et al. 2007). Other regions in the DNA, such as a part of the chloroplast trnl intron (p6-loop) and a part of the internal transcribed spacer (ITS2-region) in the nuclear DNA, has been suggested to improve amplification success and/or discriminatory power of plant DNA in environmental samples (Taberlet et al. 2007; Yao et al. 2010; Hollingsworth 2011). The chloroplast trnl UUA intron p6 loop was suggested to work well on partly degraded DNA, since the region is short and flanked by very conserved primer sites and amplifies particularly well (Taberlet et al. 2007). The discriminatory power of the whole trnl intron is approximately 67% at species level on the genbank database, while the p6-loop manages about 20%, but the discriminatory power is higher when concentrating on smaller regions, ecosystems or edible plants (Taberlet et al. 2007). Therefore, the trnl UUA intron p6 loop may be a good barcode-choice for environmental soil samples. Largely, species richness is the measure of alpha diversity used in edna and metabarcoding studies due to unreliable abundance estimates. Studies estimating species richness using barcoding, metabarcoding or metabarcoding of edna has been conducted on the following organism groups: benthic marine metazoans (Fonseca et al. 2010), benthic eukaryotic diversity (Pearman et al. 2015), species richness of nematodes from single specimen barcoding (Decaëns et al. 2016) and species mixtures (Foucher et al. 2004), arthropods in various habitats (Yang et al. 2014) and plants (Hiiesalu et al. 2012). Yang et al. (2014) found that arthropods from soil and leaf litter samples provided similar information regarding species richness and site divergence between forest and open land as malaise traps and canopy fogging samples. Fonseca et al. (2010) found a good correlation between species richness of marine metazoans measured by sediment edna and morphological identification, while Yoccoz et al. (2012) found soil edna to reflect above ground growth form diversity (forbs, graminoids and woody species). Pärtel et al. (2012) argues that soil edna are likely to provide higher estimates of species richness compared to regular above ground surveys, especially in temperate regions, due to above ground (abg) floral parts being present for a limited time in a yearly cycle, while below ground parts are relatively stable in time. This is in concordance with the findings by Hiiesalu et al. (2012), where plant abg species richness stopped increasing at a certain point of belowground diversity, while belowground species richness did not halt accordingly. Drummond et al. (2015) points out, that stable alpha diversity measures are highly dependent on sequencing depth as well as sequence similarity limits during clustering-steps in the bioinformatical pipeline. Thus, edna can be used as a measure of alpha diversity, but may not be directly comparable to regularly measured abg diversities. Being able to detect differences in community composition is another important feature in the quest to measure biodiversity by metabarcoding or edna. (Ji et al. 2013) showed that it was possible to use metabarcoding to recreate the pattern found using standard identification methods for trapped arthropods. Drummond et al. (2015) found the beta diversity of usual surveys of invertebrates, understory plants and trees to correlate well with data obtained from soil edna isolated with various markers. Yoccoz et al. (2012) reliably verified that trnl p6 loop soil edna can detect frloral differences along an altitudinal gradient in a similar way as an above ground survey. Furthermore, they showed that similar management decisions were reached using soil edna and above ground surveys. Thus it may be possible to differentiate communities along larger environmental gradients and soil types using soil edna. Investigations of DNA obtained from mixtures of known species suggest that sequencing of edna is a viable method for detecting species in mixed samples (eg. Hajibabaei et al. 2012; Hiiesalu et al. 2012), but in 10

11 general does not recover all species in a sample. This raises the question whether soil edna analysis is a feasible method for tracking diversity and species composition for a variety of soil environments is yet to be tested. In this study, a large scale comparison between floral soil edna and a regular above ground survey (abgs) is conducted over a range of different habitat types, spanning three of the major environmental gradients in soil moisture, successional stage and productivity. The main hypothesis is; soil edna can be used to assess floral diversity and plant compositional variation across large environmental gradients in natural habitats, using the chloroplast trnl intron p6 loop marker. To pursue this, four hypotheses were tested. Soil environmental DNA from the trnl p6 loop: 1. Detects variation in plant composition between sites similarly to an above ground survey. 2. Detects the effects of moisture, nutrients and succession on species composition similarly to an above ground survey. 3. Retrieves species, genera or families known to occur at a given site. 4. Give reliable estimates of known alpha diversity within the sites. 11

12 Methods: Collection of samples in the field Sample sites The current study is part of the project BioWide, where biodiversity is investigated at 130 sites in different habitats in Denmark. The 130 sites were situated in regional clusters (figure 2), with each region containing each of 26 strata. 18 strata were selected among natural and semi natural vegetation to include all combinations of three gradients; moisture (wet, moist, dry), productivity/nutrients (rich, poor) and successional stage (late, mid, early). In addition six strata were meant to cover examples of cultivated vegetation including; oldfield, rotational field, ley as well as beach, spruce and oak plantations. The remaining two strata were assigned to a hotspot category based on an internet vore for exceptionally species-rich sites amongst natural historians (not necessarily for plants). Each site was 40 times 40 meters and was, as far as possible, placed in a homogenous environment with regard to habitat type. Each quadrat was divided into four smaller quadrats, marked by four colored flags in each corner, and a white center-pole (figure 3). Midway between the center-marker and a given corner, a small pole of the quadrat color was placed, serving as the center of a vegetation survey 5 meter circles. Aboveground species list and environmental parameters Species lists were made between July and September 2014 by an experienced botanist and bryophyte expert. Species presence was registered in four 50 by 50 cm quadrats situated at each midway-point between the corners and the center of the sample site (see figure 3). Additional species were registered within a 5-meter documentation circle (radius 5m) centering on the 50 by 50 cm quadrats. Thus, just under 20% of the total sample sites were a part of the abg floral survey. Furthermore, a short evaluation of dominating species was made, if such a pattern was clear. Various environmental parameters were measured, including soil moisture, soil and organic matter ph, counts of trees above 40 cm DBH, light intensity, surface temperature and Figure 2: Regionwise distribution of samplesites in project BioWide. Each reagion is represented by a color category, while the clusters are represented by a unique color within each color category. Red: western Jutland. Blue: Eastern Jutland. Yellow: northern Jutland. Purple: Funen and the isles. Green: Zealand. (Google Earth 2013) distance to intensively cultivated farmland and natural areas, just to mention a few. Furthermore, Ellenberg values were calculated from the abg floral survey species lists. 12

13 Figure 3: Schematic drawing of a sample site, with each corner being marked by a colored flag, and the center being marked by a white flag. Within each quadrat, a pole with the quadrat color marks the center of the five meter documentation circles used for the above ground vegetation survey. Soil samples for edna Soil cores were collected in October and November 2014, using a weed extractor tool with a curved open blade (Wolf- Garten, iw-m ), collecting the top fifteen centimeters of soil, with most soil from the top 7 centimeters, and less from the lower 8 cm. A soil sample was taken in every intersection of a 9 by 9 grid, with approximately 4 meters between every intersection. Thus, 81 soil cores were collected at each sample site, pooled in a 15 liter plastic barrel (CurTec Wide neck drum, HDPE), and homogenized with a drilling machine (HILTI Cordless Combihammer) mounted with a mortar mixing paddle within 24 hours of soil collection. Subsequently, when the collected soil appeared to be homogenized (after at least 3 minutes of mixing), 5 subsamples of 5-10 tablespoons were stored in ziplock bags at -18 C. Between each handling of soil from a given sample site, the mixing paddle was cleaned in two sets of clean water, and the barrels were cleaned with water until no left over soil was visible. Afterwards, the barrels were shaken with 1 dl 5% bleach solution (Sodium hypochlorite) and left for an hour before rinsing it with water. The sampling design was inspired by (Taberlet et al. 2012b), and permitted the processing of 6-9 sample sites per day including transportation. DNA extraction, PCR and sequencing DNA from the soil samples were extracted in April and beginning of May 2015 at the Center for GeoGenetics, the Natural History Museum of Denmark at the University of Copenhagen. Extractions were made on approximately 4g of soil using the MoBio Power max soil isolation kit (MOBIO, Carlsbad, California, USA), using the manufacturers protocol, with a few modifications: 1) 4mL of 1M Calcium Carbonate suspension was added to ease DNA extraction, 2) the sample was shaken for 10 minutes in the tissuelyzer II (QIAGEN, Hilden, Germany) at maximum speed, and 3) the final elution step was done in three steps with 1xTE elution buffer heated to 56 degrees Celsius. A cleanup step after the primary extraction was conducted using MoBio PowerClean kit, following manufacturer s instructions, except for a dividing the elution step in three with heated elution buffer (1xTE buffer). Extracts were stored at -20 degrees celcius. DNA concentrations were measured using Qubit dsdna high sensitivity assay (Life Technologies/Thermo Fisher Scientific Waltham, Massachusetts, USA), and diluted for PCR. Amplifications of the DNA was done using 1µL of DNA extract, 0.2µL of Tag gold polymerase and dntp s, 1µL BSA, 2.5µL of Gold MgCl 2 and Gold PCR buffer, 1.5 µl of each tagged primer and water adding up to a final volume of 25µL per reaction. The thermal cycler ran according to this scheme: 95 C for 10 min; 40 cycles of 95 C for 30sec, 50 C for 30 sec, 72 C for 30 sec; 72 C for 7 min; 4 C until removed. Primers trnl-g (5 -GGGCAATCCTGAGCCAA-3 ) and trnl-h ( 5 -CCATTGAGTCTCTGCACCTATC-3 ) was used to target the P6 13

14 loop of the trnl UUA intron (Taberlet et al. 2007), and both primers were modified with a unique six basepair tag used to assign the amplified sequences to the correct sample (Schnell et al. 2015). Amplified DNA was stored at -20 C. The PCR products were cleaned and subsequently mixed in similar proportions based on Agarose gel electrophoresis band strength. An Illumina TruSeq DNA PCR-free library (Illumina Inc., San Diego, CA, USA) was built on the mixture of PCR products, followed by a minelute purification and checked on a Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). In total, three PCR replicates were built into six libraries, all of which were sequenced twice on individual Illumina MiSeq sequencing runs, using 120 bp paired-end reads in both directions. Database construction The reference database was constructed using OBITools (Boyer et al. 2016) as suggested in the OBITools tutorial ( but was built on Genbank/ncbi plant sequences, including lichen, algae and bryophytes (gbpln*seq.gz on the ftp://ftp.ncbi..nlm.nih.gov/genbank/). This is referred to as the full database. A database containing only species known to occur in Denmark was constructed using lists from the Danish website allearter.dk (DanBIF & Skipper) on vascular plants and bryophytes with accepted Danish species. The lists contain species accepted as Danish, and both mosses and vascular plant lists were used. From the species list, the scientific name was used to extract taxonomic id (taxid) from the ncbi taxonomy. The ids were then used to extract the sequences with corresponding taxid from the constructed ncbi-full database. The resulting multifasta-file comprises the Danish database. This database contain 499 sequences, some of which belong to the same species, which means that possibly less than ~450 species are present in the database. A script for the database construction workflow can be found in the supplementary materials. Raw data processing From raw data to OTU and sites by species table Raw data was processed using either OBITools (Boyer et al. 2016) or the Mahé pipeline using vsearch, swarm and cutadapt constructed primarily according to Frédéric Mahé (Mahé 2016), as described in the following paragraphs. The OBITools pipeline: the forward and reverse illumina pairedend strands were assembled using the illuminapairedend function with a minimum match of 40%, and unaligned sequences were removed. The sequences were assigned to samples using ngsfilter with an allowance for two errors for matching primers, but no errors for matching tags. Before dereplication, the sequences were modified to keep only the sample attribute. The resulting files from six sequencing libraries were concatenated and dereplicated globally using obiuniq, keeping information about sequence origin. To clean the dataset, sequences shorter than 10bp were removed, and stepwise pcr errors were attempted removed by obiclean, using an r value of Since the second sequencing run finished after the primary data processing of the first sequencing run, the fasta files from both runs were concatenated at this point, and another dereplication using obiuniq was completed. As a final denoising step OTU s being present only once in the data was removed and finally a table was constructed using the obitab function. These OTU s make out the dataset called OBITools raw. The OBITools db dataset was constructed by assigning a taxon to OTUs by using the OBITools ecotag function as described below. 14

15 The Mahé pipeline is more or less as described by Frédéric Mahé (Mahé 2016): The forward and reverse reads were assembled using Vsearch (Rognes et al. 2015) function fastq_mergepairs. To demultiplex the aligned sequences, a list of tags and primers are supplied to Cutadapt (Martin 2011), which assigns a sample to the sequences and trim the primers. Using Vsearch, unassigned sequences are removed and error rates added before running dereplication at the sample level. After concatenating the sequences from the sequencing libraries, a global dereplication followed by global clustering using Swarm was run (Mahé, Rognes, Quince, de Vargas, & Dunthorn, 2014; Tiffin & Ross-Ibarra, 2014), merging sequences differing at a single basepair in the quest to eliminate stepwise PCR errors. Before building an OTU table, a chimera checker was run, to identify possible chimera from sequencing. The resulting OTU table was filtered to only contain OTUs defined by: a count above one, not a chimera, present in at least two samples and with an expected error rate of less than The resulting OTU table makes out the Mahé raw dataset. To construct the Mahé db dataset, the OTU representatives from the mentioned filtered OTU table was assigned to a taxon using the OBITools ecotag as described below. Taxon assignation for both the OBITools db and Mahé db datasets were accomplished using the OBITools ecotag command, a reference database in fasta format containing only sequences with species known to occur in Denmark (see database construction) and a 90% similarity margin. Sequences with no assigned taxon were selected and compared to the full database with a lower limit of 90% similarity. After taxon assignation, ranked sequences resulting from comparison to the Danish database and all sequences from the comparison to the full database were concatenated and a table was made using obitab. The resulting table makes out the OBITools db dataset, but a little further processing was needed for the Mahé db dataset. Merging of the representative sequence table containing information about OTU taxonomy from the ecotag command and the table from the Mahé pipeline was done in R, along with correcting the arbitrary site names from the Mahé pipeline. Further filtering of the datasets in R (R Core Team 2013) was conducted as follows. For datasets with taxon assignation, data processing in R included discarding OTUs with no assigned taxon, as they were considered as not belonging to vascular plants or bryophytes. For all datasets, OTUs present in negative and blank samples were removed, replicates were merged and presence of an OTU was only kept if it was registered in at least two of the three replicates. Finally, sample sites with no presence of any OTU were removed, site SV089 was also deleted due to an error in the NGSfilter file, and data was converted into presence/absence in a site by species table. Furthermore, potential outliers were removed since they diminish the patterns found in the remaining data. Thus, different constellations of data were constructed: (1) Sequence data run through the OBITools pipeline in two different versions; (a) A raw version, where OTUs were not assigned to a taxon, but each was considered as a species, (b) a database version, where OTUs were compared to a reference database, and hits below 90% were removed. For each of these versions a and b, an OTU was considered as present in a sample if it was found in (i) two of the three replicated, or (ii) all three of the replicates. Thus, four versions of the data run through OBITools is presented: OBITools db 2 rep, -db 3 rep, -raw 2 rep and - raw 3 rep. Similar 4 versions is presented for the Mahé pipeline; (a) a raw version, where the OTUs were not compared to a reference databse, and (b) a version where OTUs were assigned to taxa via ecotag in obittols, and OTUs with less than 90% match in the databases were removed. Similarly, two variations of the data regarding replicates were constructed. The four versions of the Mahé pipeline presented are: 15

16 Mahé db 2 rep, -db 3 rep, -raw 2 rep and raw 3 rep. In total, 8 versions of the trnl soil edna data were created in order to observe the effect of pipeline type and replicate influence. Above ground survey species lists and environmental variables Data was processed and analyzed in R, with the assistance of the reshape package (H. Wickham 2007), where a sites by species table were produced indicating presence/absence. If sample sites were removed from analysis in the edna data (eg. if no OTUs were present after filtering), the same was done in the above ground survey in order to properly compare the two. Environmental variables measured in the sample sites were also supplied in R to make environmental fittings on ordinations and test the effect of physical and chemical parameters on OTU count. Data analysis The statistical analyses were conducted in R (R Core Team 2013), using the vegan (Oksanen et al. 2013) and MASS (Venables & Ripley 2002) packages. Nonmetric multidimensional scaling (NMDS) analyses were done using metamds with four dimensions and a maximum of 100 tries. To compare the edna dataset with the botanist collected data, the distance matrixes constructed by the vegdist (bray Curtis) function were compared by correlation using a mantel-test and the NMDS-ordinations were compared using a Procrustes test (function mantel and protest in vegan). Graphics were constructed using the R basics, vegan ordiplot and the ggplot2 package (Wickham 2009), and environmental variables were fitted to the ordinations using the vegan envfit function. Correlations were done using cor.test in R. Diversity estimates Due to issues regarding variance in sampling depth, the alpha diversity in the form of species richness was calculated based on three versions of the data, one where the samples where to only contain reads corresponding to the fifth quantile of the read count, one with reads corresponding to the sample with the lowest read count, and one where uneven sequencing depth was not accounted for. The resampling was done on the OTU-table using the rarefy_even_depth function from the R package phyloseq (McMurdie & Holmes 2015). The alpha diversity was calculated as the sum of OTUs in a sample in the edna data, and the sum of species in the abg survey. The two edna species richness estimates were checked for correlations to the abg survey estimate using cor.test from R basics. Recovered taxa Taxa lists were extracted from the sample sites within the Silkeborg cluster. The species list from the abg survey was assumed to represent the correct or expected taxa at a given site, thus representing a list of taxa which were likely to be found in the soil edna (see the discussion for more details on this issue). Hit and error scores were calculated for the species, genus and family level respectively. A hit was defined as the presence of the given taxon in both the edna and the abg survey taxa lists. For each taxon, the matching genus and family was retrieved, to allow comparison of taxa lists at genus and family level. For OTUs assigned to a tribe the matching family name was retrieved, since no comparison at tribal level was done. Duplicates in the species lists were removed, to compare only presence/absence of any given taxon. Taxa that may represent false negatives has been marked in the species lists (table 2 and supplementary materials tables S2-S9). 16

17 Results Ordinations The data filtering of the OBITools raw datasets resulted in and 5242 OTUs for the two and three replicate versions respectively. After taxon assignment, and removal of OTUs with no match in the database, the OBITools database two and three replicates versions yielded 3822 and 2006 OTUs respectively. In comparison, the abg dataset consisted of 871 species. Despite large variation in OTU counts, the comparisons between the abg survey data and all OBITools datasets with respect to compositional variation yielded highly consistent results, not proving significant differences with a p-value of for both the mantel test and the protest. Correlation coefficients and r 2 values can be seen in table 1 and figure 4, with r 2 values varying between for the mantel test and for the protest correlation. NMDS ordinations of the abg survey and OBITools edna data also appear similar when projected (figure 5 and supplementary materials figure S4-S). Note that the axes have been inverted and swapped in some illustrations. name Replicates count R R 2 p m12 2 corr R 2 p OTU Mantel Mantel Mantel Protest Protest Protest Protest sites obi-raw ,6552 0,4293 0,001 0,2763 0,8507 0,7237 0,001 obi-raw ,5827 0,3395 0,001 0,4024 0,7730 0,5975 0,001 obi-db ,5899 0,3480 0,001 0,3349 0,8155 0,6650 0,001 obi-db ,5361 0,2874 0,001 0,4145 0,7652 0,5855 0,001 mahe-raw ,5899 0,3480 0,001 0,3369 0,8143 0,6631 0,001 mahe-raw ,4934 0,2434 0,001 0,5279 0,6943 0,4821 0,001 mahe-db ,5359 0,2872 0,001 0,4382 0,7495 0,5618 0,001 mahe-db ,4853 0,2355 0,001 0,9871 0,1137 0,0129 0,123 mahe-db-3- out # ,4857 0,2359 0,001 0,5739 0,6528 0,4261 0,001 Table 1: The test results from comparisons of the different edna datasets with the abg survey dataset. The only variation of the edna data returning significantly different from the abg survey data, is the Mahé database 3 replicates version, and only so when comparing the NMDS ordinations (protest), not when comparing distance calculations (Mantel-test). # An outlier has been removed from the dataset. 17

18 R 2 coeficient OTU count 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 Ordination and distance matrix correlations Mantel R2 Protest R2 Final OTU count Figure 4: Graphical representation of the R squared values for both Mantel and Protest correlations between the abg survey dataset and each of the edna datasets. OTU count from the edna data has been added, and should be read on the second y-axis. All datasets, except for the Mahé-db-3replicates, yield Protest R 2 values above 0.42, and also stands out as the only pipeline version being significantly different from the abg survey data, when looking at the ordination comparison (Protest). For the data processed by the Mahé pipeline, the final OTU count in the analyses was 3530 and 1432 for the Mahé raw two replicates and three replicate versions respectively. The Mahé db datasets yielded 1358 and 767 OTUs respectively for the two and three replicate versions. When comparing to the abg survey dataset, all Mahé datasets proved to not be significantly different (table 1), with the exception of the Mahé db 3 replicate data, which returned a protest significance of 0.123, but a mantel significance of Looking at the NMDS ordination, a very clear outlier was present, making the majority of the sample sites cluster tightly. Removal of the outlier returned the significance of the correlation between the edna data ordination and the abg survey ordination, albeit with the lowest protest correlation in the peloton. It is also worth noticing that the Mahé-db 3 replicates version of the edna data renders only 119 or 118 sample sites in the analysis, out of the total 129 sites in most of the other analyses (table 1), since the sequence filtering left these sites with no OTUs present. An overview of determination coefficients can be seen on figure 4, where the final OTU count is also plotted. The NMDS ordination plots from the Mahé variations of the edna data, also comes across very similar when examining the ordinations visually (figure 5), but with the 2-replicate versions giving a much clearer picture. Outliers seem to become more frequent in the 3-replicate versions, and in the Mahé db 3-replicate dataset, an extreme outlier was removed in order to provide a pattern at all. The resulting pattern still contained three to four outliers, but removing them did not make it more or less clear. Note that the axes of the edna ordinations in figure 5 have been swapped and/or inversed to make the similarity with the abgs ordination more visible. 18

19 Figure 5: Non-metric multidimensional scaling (NMDS) ordinations showing the two primary axes of different versions of the edna data and the corresponding ordination of the abg survey data. To be able to compare the ordinations visually, the points have been assigned to the sample sites predefined stratum type. The upper graphs (OBITools raw 2 replicate version) represent the ordinations with the highest protest correlation, while the bottom graphs (Mahé db 3 replicates) represent the lowest protest correlations. For the Mahé db 3 replicates ordination, the extreme outlier has been removed note that subtle outliers are still present. Overall, all pipeline data versions was not significantly different from the abg survey in community composition, with determination coefficients above The two datasets with the lowest determination coefficients both suffered from reductions in analyzed sample sites (see table 1) due to no presence of OTUs in the given sites after data processing. All versions of data having been filtered using the Mahé pipeline shows lower correlation with the abg survey data, than the corresponding version in the OBITools pipeline. 19

20 Figure 6: NMDS ordination axes with the highest r 2 values for the given environmental gradients shown for the OBITools raw 2 replicates data and the corresponding abg survey data. The soil moisture graphs are 20

21 colored according to measured soil moisture, the nutrient graphs are colored according to Ellenberg N and the succession graphs are colored according to light intensity. Grey points are NA values. R 2 values and significance for the fittings: edna) MedianSoilM = ***, LeafN = ***, LeafP = ***, EllenbergN = ***, MeanIntensity_day = ***, TreesM40DBH_count = ***. Abgs) MedianSoilM = ***, LeafN = ***, LeafP = ***, EllenbergN = ***, MeanIntensity_day = ***, TreesM40DBH_count = ***. The edna ordinations resolve the environmental gradients in a similar fashion as the abg survey does. Environmental fittings on the abgs NMDS ordinations, here exemplified by values are for OBITools raw 2 replicates, showed very high r 2 -values for median soil moisture (0.6952), and ellenberg N (0.8774), moderate to high values for mean light intensity (0.5262), tree counts at 40 cm DBH (0.4412) and leaf N (0.3963), while leaf P only signifies a poor correlation (0.2659)(see figure 6, graphs to the right). Ellenberg N are expected to be high in the abg dataset, since it was calculated based on plants from the abg survey. For the edna datasets, the strength of the environmental fittings varied. For median soil moisture r 2 -values ranges between and , with the Mahé db datasets showing poor correlations, while the Mahé raw versions showed moderate or good correlations. All OBITools pipeline variations showed very good determination coefficients above Leaf N and P showed generally poor correlations with r 2 -values between for leaf N and for leaf P. Ellenberg N was markedly better correlated with determination coefficients between Light intensity and count of trees above 40 cm DBH showed moderate correlations with r 2 -values ranging from for light intensity and for count of trees. The OBITools versions in general showed higher correlations, while the Mahé pipelinse in general show lower correlations with a given environmental parameter. Environmental fitting values can be found in the supplementary material table S1. Taxonomic assignment The taxa recognition in the datasets where the OTU sequences were compared to the Danish- and full reference databases varied in both specificity and estimated error rate. On figure 7 it is evident that the percentage of correctly identified taxa increases with taxonomic level, but varies between edna data pipeline versions from 2.2 9% at species level, % at genus level and 12.3 to 35,7% at the family level. 21

22 Percentage taxons identified Average identification rate across nine sample sites Mahé 3 rep Mahé 2 rep OBITools 3 rep OBITools 2 rep species edna / species abgs genera edna / genera abgs families edna / families abgs Figure 7: Graphical representation of the average ID rate with standard error from the nine sample sites in the Silkeborg cluster. The red bars are species identification rates, while the purple and blue bars are average genus and family identification rates respectively. The pattern displayed in the legend shows which edna dataset the bars represent. The average number of hits at the species level for the Mahé 3 replicates pipeline was 3.33, which was also the case for hits at the family level, while genus identification number was the lowest at all with 2.22 hits (see figure 8). However, this pipeline only produced errors at genus level. The number of positive identifications increased with the 2 replicate version of the Mahé pipeline, with 6.11 correcly identified species and families, while genus identifcation was lower, with only 4.11 genusses on average. For this pipeline, errors were also most pronounced at the genus level, but only 1.6 on average. Almost similar identification numbers were found for the OBITools 3 replicates pipeline, but with higher error counts than the Mahé pipelines at all taxonomic levels. The pipeline with the most positive identifications on average was the OBITools 2 replictes version, hitting the highest number at all taxonomic levels. However, as can be seen in figure 8, this pipeline also accounts for the most errors across all taxonomic levels. 22

23 Average number of hits or errors Average hits at taxonomic levels hits Mahé 3 rep errors Mahé 3 rep hits Mahé 2 rep errors Mahé 2 rep hits OBITools 3 rep errors OBITools 3 rep hits OBITools 2 rep errors OBITools 2 rep Species level Genus level Family level Figure 8: The average number of hits or errors with standard error shown at the given taxonomic level. The shaded bars represent the average error counts, while the solid bars represent the average taxon count for the specified edna dataset. The taxa list from Letmose (sample site ES063) is supplied in table 2 as an expample of which taxa were retrieved at the current site. Two of the dominating genera were retrieved (Sphagnum and Eriophorium), while Molinia only was represented at the family level. However, Phragmites australis, a sister-genus to Molinia was found in all edna versions. All retrieved taxa are likely to be found in Denmark, but species not verified by the abgs are frequent and considered as errors (unmarked or red in table 2). Errors can be devided into errors caused by the lack of the sequence in the restricted database and potentially false negatives, ie. species that are likely present, but remains undetected by the abg survey. Most errors in this study are likely to be a result of the incomplete coverage of the sequences in the restricted and/or full database. The species list from sample site ES063 Letmose can be found in table 2, while the remaining lists from the Silkeborg cluster can be found in the supplementary materials (tables S2-S9, sample sites: ES062 Urfuglebakke, ES064 Mossø, ES065 Tørvefladen, ES066 Odderholm, ES067 Højkol/Rye sønderskov, ES068 Haarup Sandeand, ES069 Knagerne and ES070 Gjessøvej ). 23

24 ES063 Letmose Mahe 3rep: Abg survey Species Genus Family Higher taxa Aulacomnium palustre Eriophorum angustifolium Eriophorum Cyperaceae Calluna vulgaris Phragmites australis Phragmites Poaceae Carex nigra Mahe 2rep: Deschampsia flexuosa Species Genus Family Higher taxa Drosera rotundifolia Eriophorum angustifolium Eriophorum Cyperaceae Dryopteris carthusiana Phragmites australis Phragmites Cariceae Empetrum nigrum Poaceae Erica tetralix Obitools 3rep: Eriophorum angustifolium Species Genus Family Higher taxa Eriophorum vaginatum Isothecium myosuroides Sphagnum Cyperaceae Hypnum jutlandicum Carex pallescens Isothecium Sphagnaceae Leucobryum glaucum Calluna vulgaris Carex Lembophyllaceae Molinia caerulea Eriophorum angustifolium Calluna Ericaceae Picea sitchensis Sphagnum subsecundum Eriophorum Poaceae Pleurozium schreberi Phragmites australis Phragmites Polytrichum commune Sphagnum fimbriatum Echinochloa Polytrichum strictum Echinochloa crus-galli Potentilla erecta Obitools 2rep: Sphagnum cuspidatum Species Genus Family Higher taxa Sphagnum fallax Salix triandra Sphagnum Pooideae Bryidae Sphagnum fimbriatum Trifolium repens Carex Poaceae Sphagnum palustre Isothecium myosuroides Salix Paniceae Sphagnum papillosum Carex pallescens Trifolium Cyperaceae Sphagnum subnitens Calluna vulgaris Isothecium Sphagnaceae Vaccinium oxycoccos Fagus sylvatica Calluna Lembophyllaceae Vaccinium vitis-idaea Carex maritima Fagus Ericaceae Eriophorum angustifolium Eriophorum Salicaceae Sphagnum subsecundum Phragmites Fabaceae Phragmites australis Echinochloa Fagaceae Sphagnum fimbriatum Echinochloa crus-galli Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. Table 2: Species lists from sample site Letmose ES063. The genus column contain both direct genus assigments from the ecotag process, as well as the matching genus for the species allocated by the ecotag function. The same accounts for the family column, that also contains the matching family for the genera, as well as direct family assignments by the ecotag database assignment tool. The meaning of the colored 24

25 shadings are described in the bottom of the table. The abg species list can be seen to the right, while the edna taxa lists are found to the left. Diversity estimates The edna data was checked for even sampling depth, and for all versions of the data, a significant correlation was found between the sequence read count and the OTU count for the sample sites (see figure GG and supplementary materials figure S3). To account for this correlation, and even out sampling depth, the data was to the 5 th quantile of the original read count or to the minimum read count observed in a sample, yielding the rarefaction plots visible in figure 9. The evening out of read count results in a reduction in OTU count as can be seen on the y axes. Some of the data versions suffered large reductions in OTU count, here exemplified by the Mahé db 2 replicate version reducing maximum OTU count from just under 120 to 25 and 2 in the and minimum versions. The OBITools pipelines, exemplified by the OBITools db 3 replicates data, only suffered reduced from a maximum of 140 to 120 and 35 OTUs. 25

26 OTU count OTU count OTU count OTU count OTU count OTU count Not Resampled Resampled minimum Read count/1000 Read count/1000 Read count/1000 Not Resampled Resampled minimum Read count/1000 Read count/1000 Read count/1000 Figure 9: Rarefaction plot showing the sequence read count per sample versus the sample OTU count before and after resampling of the dataset. Each dot represents one sample site. A & B: the OBITools db 3 replicates data before and after resampling and C & D are the same for the Mahé db 2 replicates. There is a positive correlation between the number of sequence reads and OTU counts (Pearsons cor.test in R basics: Mahé db 2 replicates: cor= ,p=2.084e-08, r 2 = OBITools db 3 replicates: cor= , p=1.266e-05,r 2 = ), indicating a divergence in sampling depth between samples

27 Coefficients of determination The alpha diversity found in the abg survey and edna data was highly significantly correlated, both when looking at the original and the 5 th quantile data, and for some minimum data versions (figure 11, table 3 and figure 10). Correlations values and further examples of scatter plots can be found in the supplementary materials figure S3. When looking at figure 10, it is clear that for the Mahé pipeline resampling of the data results in a much less correlated alpha diversity measure (still p=<0.001), while the OBITools pipeline data yielded similar or slightly improved correlations with the abg suvey species richness. However, determination coefficients are generally low, as also indicated by the large amount of scatter on the correlation plots (figure 11). 0,35 0,3 Determination coefficient for alpha diversity comparisons Not Resampled 5th quantile Resampled minimum 0,25 0,2 0,15 0,1 0,05 0 Figure 10: Graphical representation of the Pearson coefficients of determination (R 2 ) from the comparison between the edna OTU and abg survey species richness. See table 3 for significance of the individual correlations. For the OBITools pipelines, the amount of explained variance does not change much from the original to the 5 th quantile data and only the database versions change drastically when the minimum resampling was employed. For the Mahé Pipelines, the species richness from both data versions explains markedly less of the variation found between the edna and abg survey species richness. However, in general less than 32% of the variation between abg species richness and edna OTU richness has been explained. 27

28 Species in abg survey Species in abg survey Species in abg survey Species in abg survey Species in abg survey Species in abg survey Not Resampled Resampled minimum OTU count OTU count OTU count Not Resampled Resampled minimum OTU count OTU count OTU count Figure 11: Scatter plots of taxonomic richness in OTU number from edna data and species richness from the abg survey data. The plots on the left are the original data before resampling, the middle plots are the data, whilethe right plot are data to the minimum read count. All plots present a significant correlation, values can be found in table 3 Note the reduction in OTU count when resampling the data.

29 Not t df confidence low high p correlation original R 2 OBITools db 2rep E OBITools raw 2rep E OBITools db 3rep E OBITools raw 3rep E Mahé db 2rep E Mahé raw 2rep E Mahé db 3rep E Mahé raw 3rep E Resampled 5th quantile t df confidence low high p correlation R 2 OBITools db 2rep E OBITools raw 2rep E OBITools db 3rep E OBITools raw 3rep E Mahé db 2rep E Mahé raw 2rep E Mahé db 3rep E Mahé raw 3rep E Resampled minimum t df confidence low high p correlation R 2 OBITools db 2rep OBITools raw 2rep E OBITools db 3rep OBITools raw 3rep E Mahé db 2rep Mahé raw 2rep Mahé db 3rep E Mahé raw 3rep Table 3: Pearson correlations between the edna dataset OTU count and the abg survey species richness for the original data and two data versions. OTU counts were used as an estimate for species richness in the edna data, while observed species richness was used for the abgs data. Original and 5 th quantile datasets yield highly significant results, but with low coefficients of determination. For the minimum version, only some pipeline versions yielded significant correlations. The influence of physical and chemical parameters To test whether the difference found in species richness between the two survey types were affected by chemical or physical parameters measured in the plots, it was tested whether the residuals extracted from a linear model correlated with soil ph, organic matter ph, mean surface temperature and mean soil moisture. Some pipeline versions yielded significant correlations (in either original, or minimum versions) with some of the physical and chemical parameters, but no clear pattern was found (supplementary materials tables S14-S15). Similarly, these parameters were tested against the OTU count in all samples to check whether edna species richness estimates was directly affected by temperature, ph and moisture content of the sample sites. In some pipeline versions a correlation was found when testing soil ph, organic matter ph and surface temperature against OTU count, both in cases where the data had been to the minimum read count, and in cases where no resampling had been done. However, correlation coefficients were low. To check whether this only applied to the edna data, similar correlations were made for abg species richness. Soil moisture and both measures of ph returned significant correlations with abg species richness, but with 29

30 very low determination coefficients. Correlation results can be found in the supplementary materials figure S6-S9 and table S10-S13. Discussion In summary, edna metabarcoding of plants using the p6 loop of the trnl region reproduces similar compositional patterns as a regular above ground survey, irrespective of the pipeline methods used to treat the raw sequence data, and in this analysis for both ordinations and mantel tests. The method also distribute the sample sites along the three natural gradients over which the sample sites were placed; nutrients, succession and moisture. The alpha diversity measured as OTU count from the edna trnl data correlated significantly with the species richness from the abg survey, despite a high amount of scatter on the scatter plots (figure HH) and very low coefficients of determination (table 3). The highest percentage of retrieved taxa known to be present in the Silkeborg Cluster was 35.7% at the family level, and lower for the genus and species level. So named errors in taxon assignation was common for some versions of the edna data, but is likely to be the result of reference database issues as discussed later. Two of the four sub-hypotheses can be reliably confirmed by this study. Data from sequencing of trnl p6 loop soil edna is capable of detecting floral site divergence similarly to an aboveground survey, as shown by ordinations and protest correlations. The moisture, succession and nutrient gradients underlining the study sites were also retrieved from the edna data, and fittings were comparable to the abg survey for the best correlated datasets. However, most species, genera or families were not recovered reliably in the current study, despite the use of a restricted reference database for sequence comparison. Despite highly significant correlations between alpha diversity estimated by edna and abgs, a reliable estimate of species richness with high explanatory coefficients was not reached using soil trnl p6 loop edna in this study. Thus, soil edna may be used to detect floral changes and site divergence, but obtaining reliable estimates of species richness remain a challenge for future studies. ID rate Up to 9% of species known to exist in the specific sample sites in this study were recovered, a percentage that increased with taxonomic level, thus up to 35% of families were recovered. Estimations of species or taxa retrieveability from environmental soil samples are few, but are relatively high for water samples (100% of known species, Thomsen et al. 2012b), however, Yoccoz et al. (2012) recovered up to an assumed 33% of plant species in tropical soil edna using the trnl p-loop. Taxa retrieveability in general has been tested on laboratory mixtures of known species; Porazinska et al. (2009) recovered more than 90% of known nematode species in sample mix, while Hajibabaei et al. (2012) recovered up to 90% of all benthic organism species, and 100% with an abundance above 1% of the total mixture. A similar abundance relation was found by Zhan et al. (2013) on aquatic invertebrate larvae, while varying proportions of species were unrecovered in known species mixtures of roots (20%, Hiiesalu et al. 2012), arthropods (24%, Yu et al. 2012), diatoms (up to 24%, Kermarrec et al. 2013) and insects (3-8% of known OTUs, Zhou et al. 2013). However, these studies focused on laboratory mixes of known species. Environmental samples are more complex, and recovery rates may therefore be considerably lower, but still be abundance dependent. The sampling design for this experiment included a reduction in sampling material from 81 soil cores per sample site to 4 grams of soil for extraction, returning 5mL of crude DNA extract of which 1 µl was amplified and used for sequencing. Thus, not all taxa are likely to be represented in the soil samples, DNA extracts and 30

31 sequencing results. However, the retrievability estimates in this study are still assumed to be underestimated due to artefacts of the sampling design and problems with the database discussed later. In the current data, more taxa were assigned to species level than the previously mentioned 2-9%, but were considered as errors due to the lack of the given species in the abg survey species lists (figure 8). The assumption that the abg survey taxon lists are the unifying answers may in many cases be untrue, due to the difficulties in detecting every species present at a plot (Schmidt et al. 2013). Species that may prove difficult to find include dormant species, seedlings and seeds as well as small and low abundant species. Some of these species may be present in edna samples, though less abundant species may remain undetected using the edna metabarcoding method as well. Another feature rendering the assumption of wrongly assigned species questionable is the fact that only species being present within the four 5 meter documentation circles are registered in the abg survey. This corresponds to an area of π = 314 m2 out of the total sample site area of 40m 40m = 1600 m2, or just under 20%. The soil edna samples were taken across the whole area in a 9*9 sample grid (see the methods section), thus the edna samples may catch species that by chance are found outside the four 5-meter circles. Still, the abg survey does represent a minimum of species expected to be present in the soil DNA, whether the trnl method retrieves them or not. Thus, the taxa considered as errors in this analysis, may represent: misidentification in the database sequences resulting in the wrong taxon being assigned to the sequence, lack of the target sequence in the reference database, possible naming confusion or the low resolution of the trnl p6 loop rendering two or more identical sequences in the database belonging to different species resulting in assignation of the given sequence to a higher taxonomic level. Problems with the reference data base In the Letmose sample site all edna pipelines recovered Phragmites australis, a poaceae species not found in the abg survey (table 2). However, in the abg survey, the poaceae Molinia caerulea was abundant in the sample site. Molinia was not present in the Danish database neither as species or as genus, thus the species resembling the sequence most is likely to be Phragmites australis, a close sister-genus in the tribe molinieae, as found in the edna. It is known that the trnl region, including the p6-loop, is low in resolution within the Poaceae (Taberlet et al. 2007), which adds to the difficulties when the reference database does not contain the desired species or genus. This also highlights the issues of using a restricted database, known not to contain all the possible species/genus/families (Kvist 2013), since it may result in a type II error (Virgilio et al. 2010), eg. the assignation of a sequence to a wrong species or genus while the desired genus may be present in the larger database. Most of the error taxa in this study belong to this category, lacking a sequence in the reference database, or having only one close relative in the database (eg. Phragmites australis, Molinia caerulea, Salix triandra see table 2 and supplementary materials table S2-S9 for species lists). The Danish database contains 499 reference sequences, some of which belong to the same species, thus probably less than 50% of the species known from the abg survey are represented in the restricted reference database. Improving the construction method for the restricted database may alleviate this issue to some extent, since a quick search on approximately fifty species from the abg survey resulted in four species being present in the Danish database, while nine of those species were present in the full database. This indicates that the retrieval method using taxid for the species is not an ideal way to construct restricted databases. However, using only the full database as reference produces both less errors, but also less hits at species level, which is according to expectance since this database contain more species within most genera, a higher taxonomic level is assigned to a given sequence (Taberlet et al. 2007) 31

32 (Full database only ID-rates can be found in the supplementary materials figure S1-S2). Yet, having a complete restricted reference database is likely to improve performance at lower taxonomic levels, but would also increase the level of type I errors, ie. the failure of detecting the correct taxa, despite its presence in the database (Virgilio et al. 2010). Furthermore, a more complete restricted reference database may also push taxonomic assignation to higher taxonomic levels (Taberlet et al. 2007).In this experiment, the database similarity limit was set to 90% when comparing to both databases, errors such as the ones described in this paragraph could potentially be limited or avoided if a more conservative similarity limit had been used. For example, comparing the OTUs to the Danish database allowing only taxa matches above say 98%, as was used on trnl p6-loop by Quéméré et al. (2013), would probably yield a list of more certain matches, while allowing the full database to catch species belonging to another genus or family with a similarity limit of minimum 90%. Such an approach could potentially avoid species being wrongly assigned to a taxa present in the restricted Danish database such as Phragmites australis from the abovementioned example. False negatives and pipeline discrepancies The error rank of some of the taxa in the edna data, may have been given to the taxa due to the issue discussed in the above paragraph, but some errors may represent another group: the potentially false negatives. In this case, false negatives represent species found in the edna data that may actually be present, but has not been verified in the abg survey. This includes early or late present species (Hiiesalu et al. 2012), since the survey was done mid-summer, as well as dormant species (Ji et al. 2013) and larger trees situated outside the 5-meter circles, but with roots extending into the sample site. In sample site ES064 Mossø (see supplementary materials table S2-S9 for species lists), Fraxinus excelcior has been found in the edna but not in abg survey. Dropped leaves or root extension from a Fraxinus outside the sample site may have contributed to the soil edna, while rendering the tree undetected in the abg survey, thus being a case of a false negative in the abg survey. The same accounts for Populus in site ES066 Odderholm, and Quercus at ES069 Knagerne (see supplementary materials table S2-S9 for species lists). Some mosses may also remain hidden, or simply be found outside the 20% of the plot covered by the abg survey. Kesanakurti et al. (2011) found 10/29 taxa to be present only in root fragment DNA collected from soil cores, and not in an abg survey on the sample site. They suggest several possible explanations: plant dormancy, low abgs sampling effort, incomplete morphological identification of abg flora and plants situated outside the sampling area with roots extending into the sample site. Yoccoz et al. (2012) found 22 species in soil edna not registered at the site by aboveground survey, and suggest they are a part of the regional species pool. Hiiesalu et al. (2012) also detected 9 species not registered above ground in the study period, but all species had previously been observed at the site. These species undetected by abgs were ephemerals, late flowering plants, or clonal plants with large root extension. Therefore, false negatives, ie. species present in soil edna data not verified by abg survey, are likely to be species actually present at the site that are difficult to retrieve by abgs. Other interesting discoveries in the taxons found in the edna in sample site ES063 Letmose is Echinochloa crus-galli, a warmth loving poaceae species, that is becoming a pest species in open agricultural fields such as beets, vegetables (Mathiassen & Kudsk 2004) and potentially maize as experienced elsewhere in warmer regions (Bosnic & Swanton 2016). However, the presence of Echinochloa in a bog raises skepticism, and this species may also be a result of the database lacking the closely related species that the sequence in reality belongs to. 32

33 Yet another curiosity found in the taxa lists from the nine sample sites in the Silkeborg cluster, is the fact that the Mahé and OBITools pipeline yield different taxa for the same sample site in some cases. An example is from the sample site 65 Tørvefladen where Galium and Hydrocotyle was found by the Mahé pipeline, but not retrieved by the OBITools data versions. Similarly Atrichum tenellum, Calluna vulgaris and Sphagnum is found by the OBITools pipeline in the mentioned site, but not retrieved by the Mahé pipeline. This difference indicates that the treatment method of the raw data is of great importance. In this case, differences may be caused by the global clustering step in the Mahé pipeline or other differences between the filtering of the raw data in the two pipelines. Thus, the resulting lists of taxa is clearly sensitive to the data processing, as also discussed by Smith & Peay (2014). Diversity measurements The measurements of alpha diversity in the form of species richness from edna and abg survey data were highly significantly correlated, despite much scatter in the correlation plot (figure 11 & table 3). Only one edna data version OTU count showed a moderately good correlation with abg species richness: Mahé db 2 rep, but only when no resampling had been done. Thus, no verifiable estimation of species richness from trnl soil edna were found in this study. Drummond et al. (2015) also did not reach any comparable estimates of alpha diversity in soil edna between six markers (including trnl) and above ground inventories of trees and seedlings, but a few others have produced more promising results. Fonseca et al. (2010) found benthic substrate edna species richness of marine metazoans to broadly resemble that of a study using morphological characters. They standardized the data by a resampling procedure, but found that standardized and original data OTU richness did not differ statistically. They further show OTU richness to be highly dependent on the sequence similarity cut-off, but found 96% to yield richness estimates that resembled that of morphological characters the most. Species richness estimates of metabarcoded trapped arthropod species were found to correlate well with species richness assessed from morphological identification (Ji et al. 2013). Hiiesalu et al. (2012) found below ground alpha diversities of metabarcoded root extractions using the trnl marker (c and d primers) to exceed that of aboveground species richness, but all species found belowground but not detected above ground had previously been found at the study site. Thus, higher species richness estimates from belowground are likely to be expected, depending on amount of soil sampled (Hiiesalu et al. 2012), sequencing depth and similarity thresholds (Drummond et al. 2015). However, both the above mentioned studies were done on collected mixtures of individuals or roots isolated from soil, and not directly on extracted soil, and are in that respect not directly comparable to the current study, but the results from Hiiesalu et al. (2012) does hold promise, since much of the floral DNA available in the soil is likely to originate from roots. Hiiesalu et al. also touches on another important matter; the influence of the amount of soil sampled. As described earlier, the amount of soil sampled in the current study came down to 4g originating from a sample site of 40 by 40 meters. It is therefore unlikely that all plants species present in the soil has been extracted and sequenced, and a underestimation of species richness could therefore be expected. However, no clear trend towards under or overestimations was found (see figure 11 for examples of scatter plots). The low explanatory power in the alpha diversity estimates in the current study is likely to be highly affected by the method used when pooling amplified DNA samples. The rough visual estimation of DNA abundance based on band strength on an agarose gel electrophoresis is likely to be one of the causes for the high variation in sampling depth, since it may be troublesome to differentiate between bands with high and very high DNA concentrations, thus resulting in differential loading of the sample DNA for sequencing. 33

34 This issue can to some extend be alleviated through resampling of the resulting data to low sampling depth (Hughes & Hellmann 2005; McMurdie & Holmes 2014), but this process also removes variation in the data (see figure 11), variation that may represent real biological species, and in the current study it did not improve the correlation between OTU count and abg species richness. Furthermore, Drummond et al. (2015) points to deep sequencing and similarity thresholds as other important factors for achieving stable estimates of alpha diversity. Fonseca et al. (2010) also stressed that alpha diversity estimates are highly influenced by similarity thresholds used during clustering in the bioinformatical pipeline. In the current study, very subtle clustering steps were used in the OBITools pipeline, where errors were attempted removed by taking sequence abundance of species differing only by 1 bp into account during the clustering step, collapsing similar OTUs if the count of one was 25% of the count of the other (obiclean with r=0.25, see Boyer et al. 2014). In the Mahé pipeline, sequences were clustered if they differed by 1 bp, irrespective of the sequence abundance. This clearly resulted in less OTUs and therefore lower richness estimates in the Mahé pipeline, but still no reliable estimates of species richness were found in any of the pipelines. Setting the similarity threshold is difficult, and due to low but differing amounts of intraspecific variation in trnl between species (Taberlet et al. 2007), clustering at 1 bp divergence may be accurate for some taxa, but too narrow a limit for other taxa with higher intraspecific sequence variation. This could result in higher diversity estimates for some floral taxa than for others. Similarly, too broad similarity boundaries may result in the clustering of more biological species in one OTU, yielding too low estimates of diversity for these groups (Valentini et al. 2008). Thus, finding a similarity threshold in between, where the overall OTU count relates to known diversities may be the way to proceed, albeit the estimates from belowground may be larger than known from abg surveys (Hiiesalu et al. 2012). The edna species richness/otu counts were found to correlate with measures of organic matter ph, soil ph and surface temperature in some data versions, albeit with very low r squared values. Since DNA adsorbs better in acidic soils (Levy-Booth et al. 2007), the positive correlation between ph and OTU count may highlight possible difficulties extracting the DNA from acidic soils. However, this was accounted for in the extraction method, where calcium carbonate suspension was added to the samples, to aid the release of DNA from acidic soils. The fact that the abg dataset showed similar positive and relatively low correlations with both measures of ph and soil moisture (see supplementary materials table S10-S13 and figure S6-S9), led to the conclusion that DNA adhering and avoiding extraction was not the explanation for the correlation between ph and edna species richness. Rather, ph may explain some of the variation in species richness between sites in both the abg survey and the edna data. A positive relationship between floral species richness and ph is generally found in European environments (Pärtel 2002), and the relation found in this study may therefore reflect a real difference in species numbers between acidic and calcareous environments. Community composition The community compositional differences between sample sites displayed by NMDS ordinations, protest and mantel correlations were similar between edna data and the abg survey, with the best correlations from 2 replicates and raw versions data and abg data. These data versions were in general based on a higher number of OTUs as compared to the database and 3 replicate versions (figure 4), which may explain why the raw versions were better correlated with the abg survey, since more difference between sites are possible with more OTUs. Some of those differences were likely removed from the low-otu versions of the edna data, resulting in a lower correlation with the abg survey. The higher protest correlations as 34

35 compared to the Mantel correlations are likely to be a result of the dimensional reduction when employing an NMDS ordination to compare variation between sites. In the current study, the variation between sites was reduced to the four ordination axes explaining most of the variation. No such reduction in dimensions is employed for the dissimilarity matrix used as a base for Mantel test, which is likely to be the explanation for the lower mantel correlations between the edna and abg data as compared to protest correlations. Not many studies have tested soil edna samples in comparison with regular measurements of community composition, but Yoccoz et al. (2012) found abg and trnl p6 loop soil edna data to group similarly in an ordination analysis of meadow and heath plant communities from Varanger, Norway. However, while they focused on two habitat types, the current study sampled much larger gradients, spanning many habitats and soil types. Drummond et al. (2015) found soil edna to predict abg beta diversity or community compositional variation similarly to traditionally used methods for sampling invertebrates as well as trees and understory plants across an elevation gradient. They tested different markers, including trnl, which correlated significantly with datasets for understory plants collected above ground. Curiously enough they found no correlation between soil trnl and an above ground tree dataset, while 18S and COI soil edna correlated with both above ground datasets. Since soil contains both tree and understory DNA, a combined dataset including both groups may give a clearer picture as found by the current study, which relied on a detailed abg survey of vascular plants and bryophytes. In the current study three large environmental gradients were underlining the sampling sites; nutrients, succession and moisture. In figure 6, measured or calculated environmental parameters representing the gradients have been fitted to the NMDS ordinations, and both survey types successfully detects the effects of soil moisture, successional stage and nutrient on the species composition. The studies mentioned above, ie. Yoccoz et al., Drummond et al. focused on a single underlying gradient or whether two different habitat type communities could be separated by edna. The current study shows that data based on collected edna are able to differentiate community variation across three large environmental gradients, in a way comparable to a common abg survey. Marker choice The trnl p-loop marker used in this study has several advantages including well conserved primer sites, part of a well-studied region in the DNA, good amplification and short in length, however, the resolution is less than desired for well-studied ecosystems (Taberlet et al. 2007),, as was also the conclusion reached in the current study. For use in biodiversity measurements, knowledge about the species inhabiting the area in focus is important in order to detect invasive, rare or endangered species (Porco et al. 2013; Wilcox et al. 2013). For this purpose, the trnl p-loop may not be the metabarcode of choice, due to the low taxonomic resolution of the marker. In less described ecosystems, such as many tropical areas, this marker may prove to be very useful, mainly owing to the small size and being retrievable also in environments with quick degradation (Yoccoz et al. 2012), but also since the resolution is less problematic, because often no specieslevel reference database is available for these ecosystems (Navarro et al. 2010). However, for monitoring environmental change and possibly species richness, the trnl p-loop may prove to be adequate, or even preferable, due to the good amplification ability and the short length. Nevertheless, other markers are available if an improvement in resolution is desired. The internal transcribed spacer 2 in the nuclear ribosomal DNA has been suggested as a metabarcode for plants (Yao et al. 2010), and may prove to be useful for good taxonomic resolution in soil edna; Chen et al. (2010) found a 92% discrimination among 4800 species of medicinal plants, while Yao et al. (2010) found that it could separate 74-88% on Genbank plant sequences and further discuss the abilities and issues regarding the use of ITS as a metabarcode. 35

36 Thus, ITS2 may be suitable for soil environmental species discrimination within plants in future studies, especially in well-studied environments, as an alternative to trnl, rbcl and matk. Conclusive remarks and future studies For measuring community composition, the most suitable data versions in the current study are the raw and relatively unfiltered versions with many OTUs. Whether this is the case for data used in taxonomic assignments and species richness estimates remains unclear in the current study, and more conservartive approaches may be desirable to undertake these tasks. For both taxonomic retrieval and species richness estimates from edna, the rare species represent the largest challenge, since they are more likely to be missed: in the sampling procedure, during extraction, not being present in the amount of DNA used for PCR or discareded during error filtering in the bioinformatical pipeline. Both Smith & Peay (2014) and Drummond et al. (2015) points to deep sequencing and not PCR replicates as important in order to retrieve the most taxa, but also refers to the bioinformatical pipeline as important in order to remove errors generated from PCR or sequencing. As sequencing depth may be important, so is the amount of sampled soil. Hiiesalu et al. (2012) found an increase in the amount of species found both above and below ground with a higher amount of soil sampled for below ground analysis or area covered for abg survey. Thus, for the retrieval of species richness or more taxa including the rare ones, it may be necessary to increase the amount of soil extracted (eg. more than 4g per 40 by 40 m), or decrease the area surveyed above ground. However, more studies are needed to accurately recover the expected taxa and obtain reliable species richness estimates, focusing on obtaining equal sampling size, further analyzing the effect of a deeper sequencing (despite the high sequencing depth for the Illumina Miseq platform) and developing an easy sampling design while still processing enough soil allowing for even low abundance taxa to be recovered. From a management perspective, the use of trnl p6 loop soil edna to monitor floral biodiversity in wellknown ecosystems is unprofitable due to the relatively high cost and low species level resolution. Thus, in order to use edna to accurately measure biodiversity, improvements are needed to obtain reliable species richness estimates and other markers should be investigated for species level recoverability. 36

37 References Andersen K, Bird KL, Rasmussen M et al. (2012) Meta-barcoding of dirt DNA from soil reflects vertebrate biodiversity. Molecular Ecology, 21, Bienert F, De Danieli S, Miquel C et al. (2012) Tracking earthworm communities from soil DNA. Molecular Ecology, 21, Bosnic AC, Swanton CJ (2016) Influence of Barnyardgrass (Echinochloa crus-galli) Time of Emergence and Density on Corn (Zea mays). Weed Science, 45, Boyer F, Mercier C, Bonin A et al. (2016) obitools: A unix-inspired software package for DNA metabarcoding. Molecular Ecology Resources, 16, Buée M, Reich M, Murat C et al. (2009) 454 Pyrosequencing analyses of forest soils reveal an unexpectedly high fungal diversity. New Phytologist, 184, Butchart SHM, Akcakaya HR, Kennedy E, Hilton-Taylor C (2006) Biodiversity indicators based on trends in conservation status: Strengths of the IUCN red list index. Conservation Biology, 20, Butchart SHM, Walpole M, Collen B et al. (2011) Global Biodiversity : Indicators of Recent Declines. Science, 1164, CBOL Plant Working Group, Hollingsworth PM, Forrest LL et al. (2009) A DNA barcode for land plants. Proceedings of the National Academy of Sciences of the United States of America, 106, Chariton A (2012) Short and informative DNA products to indirectly measure vascular plant biodiversity. Molecular Ecology, 21, Chen S, Yao H, Han J et al. (2010) Validation of the ITS2 region as a novel DNA barcode for identifying medicinal plant species. PLoS ONE, 5, 1 8. Coissac E, Riaz T, Puillandre N (2012) Bioinformatic challenges for DNA metabarcoding of plants and animals. Molecular Ecology, 21, DanBIF, Skipper L Allearter.dk. Danish Biodiversity Information Facility c/o Statens Naturhistoriske Museum, Københavns Universitet, Universitetsparken 15,2100 København Ø. Decaëns T, Porco D, James SW et al. (2016) DNA barcoding reveals diversity patterns of earthworm communities in remote tropical forests of French Guiana. Soil Biology and Biochemistry, 92, Dees M, Straub C, Koch B (2012) Can biodiversity study benefit from information on the vertical structure of forests? Utility of LiDAR remote sensing. Current Science, 102, Drummond AJ, Newcomb RD, Buckley TR et al. (2015) Evaluating a multigene environmental DNA approach for biodiversity assessment. GigaScience, 4, 46. Epp LS, Boessenkool S, Bellemain EP et al. (2012) New environmental metabarcodes for analysing soil DNA: Potential for studying past and present ecosystems. Molecular Ecology, 21, F Sanger, AR Coulson (1975) A Rapid Method for Determining Sequences in DNA by Primed Synthesis with DNA Polymerase. Journal of Molecular Biology, 94, Ficetola GF, Miaud C, Pompanon FO, Taberlet P (2008) Species detection using environmental DNA from water samples. Biol. Lett, 4, Fonseca VG, Carvalho GR, Sung W et al. (2010) Second-generation environmental sequencing unmasks marine metazoan biodiversity. Nature communications, 1, 98. Foucher ALJL, Bongers T, Noble LR, Wilson MJ (2004) Assessment of nematode biodiversity using DGGE of 18S rdna following extraction of nematodes from soil. Soil Biology and Biochemistry, 36, Gebhard F, Smalla K (1999) Monitoring field releases of genetically modified sugar beets for persistence of transgenic plant DNA and horizontal gene transfer. FEMS Microbiology Ecology, 28, Giovannoni SJ, Britschgi TB, Moyer CL, Field KG (1990) Genetic diversity in Sargasso sea bacterioplankton. Nature, 345, Google Earth (2013) Denmark - the Biowide sample sites. 58* ''N 7*38'40.18''E to 54*16'54.33''N 12*33'57.35''E, Accessed H. Wickham (2007) Reshaping data with the reshape package. Journal of Statistical Software, 21. Hajibabaei M, Spall JL, Shokralla S, van Konynenburg S (2012) Assessing biodiversity of a freshwater benthic 37

38 macroinvertebrate community through non-destructive environmental barcoding of DNA from preservative ethanol. BMC Ecology, 12, 28. Hiiesalu I, Öpik M, Metsis M et al. (2012) Plant species richness belowground: Higher richness and new patterns revealed by next-generation sequencing. Molecular Ecology, 21, Hollingsworth PM (2011) Refining the DNA barcode for land plants. Proceedings of the National Academy of Sciences, 108, Hollingsworth PM, Graham SW, Little DP (2011) Choosing and using a plant DNA barcode. PLoS ONE, 6. Hughes JB, Hellmann JJ (2005) The application of rarefaction techniques to molecular inventories of microbial diversity. Methods in Enzymology, 397, Illumina (2016) Questions & Answers., How many reads can I expect from an optimally clus. Ji Y, Ashton L, Pedley SM et al. (2013) Reliable, verifiable and efficient monitoring of biodiversity via metabarcoding. Ecology Letters, 16, Kermarrec L, Franc A, Rimet F et al. (2013) Next-generation sequencing to inventory taxonomic diversity in eukaryotic communities: A test for freshwater diatoms. Molecular Ecology Resources, 13, Kesanakurti PR, Fazekas AJ, Burgess KS et al. (2011) Spatial patterns of plant diversity below-ground as revealed by DNA barcoding. Molecular Ecology, 20, Kvist S (2013) Barcoding in the dark?: A critical view of the sufficiency of zoological DNA barcoding databases and a plea for broader integration of taxonomic knowledge. Molecular Phylogenetics and Evolution, 69, Leutner BF, Reineking B, Müller J et al. (2012) Modelling forest α-diversity and floristic composition - on the added value of LiDAR plus hyperspectral remote sensing. Remote Sensing, 4, Levy-Booth DJ, Campbell RG, Gulden RH et al. (2007) Cycling of extracellular DNA in the soil environment. Soil Biology and Biochemistry, 39, Little DP (2014) A DNA mini-barcode for land plants. Molecular Ecology Resources, 14, Lodge DM, Turner CR, Jerde CL et al. (2012) Conservation in a cup of water: Estimating biodiversity and population abundance from environmental DNA. Molecular Ecology, 21, Mahé F (2016) Fred s metabarcoding pipeline. Mahé F, Rognes T, Quince C, de Vargas C, Dunthorn M (2014) Swarm: robust and fast clustering method for amplicon-based studies. PeerJ, 2, e593. Martin M (2011) Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal, 17, 10. Mathiassen SK, Kudsk P (2004) Hanespore og grøn skærmaks - biologi og bekæmpelse. McMahon SM, Harrison SP, Armbruster WS et al. (2011) Improving assessment and modelling of climate change impacts on global terrestrial biodiversity. Trends in Ecology and Evolution, 26, McMurdie PJ, Holmes S (2014) Waste Not, Want Not: Why Rarefying Microbiome Data Is Inadmissible. PLoS Computational Biology, 10. McMurdie PJ, Holmes S (2015) Shiny-phyloseq: Web application for interactive microbiome analysis with provenance tracking. Bioinformatics, 31, Navarro SP, Jurado-Rivera JA, GÓmez-Zurita J, Lyal CHC, Vogler AP (2010) DNA profiling of host-herbivore interactions in tropical forests. Ecological Entomology, 35, Nicholson E, Collen B, Barausse A et al. (2012) Making robust policy decisions using global biodiversity indicators. PLoS ONE, 7. Oksanen J, Blanchet FG, Kindt R et al. (2013) vegan: Community Ecology Package. Pearman JK, Irigoien X, Carvalho S (2015) Extracellular DNA amplicon sequencing reveals high levels of benthic eukaryotic diversity in the central Red Sea. Marine Genomics, 26, Porazinska DL, Giblin-Davis RM, Faller L et al. (2009) Evaluating high-throughput sequencing as a method for metagenomic analysis of nematode diversity. Molecular Ecology Resources, 9, Porco D, Decaëns T, Deharveng L et al. (2013) Biological invasions in soil: DNA barcoding as a monitoring tool in a multiple taxa survey targeting European earthworms and springtails in North America. Biological Invasions, 15,

39 Pärtel M (2002) Local Plant Diversity Patterns and Evolutionary History at the Regional Scale. America, 83, Pärtel M, Hiiesalu I, Öpik M, Wilson SD (2012) Below-ground plant species richness: New insights from DNAbased methods. Functional Ecology, 26, Quayle JF, Ramsay LR (2005) Conservation status as a biodiversity trend indicator: Recommendations from a decade of listing species at risk in British Columbia. Conservation Biology, 19, Quéméré E, Hibert F, Miquel C et al. (2013) A DNA Metabarcoding Study of a Primate Dietary Diversity and Plasticity across Its Entire Fragmented Range. PLoS ONE, 8. R Core Team (2013) R: A language and environment for statistical computing (R Foundation for Statistical Computing, Ed,). Vienna, Austria. Rodrigues ASL, Pilgrim JD, Lamoreux JF, Hoffmann M, Brooks TM (2006) The value of the IUCN Red List for conservation. Trends in Ecology and Evolution, 21, Rognes T, Mahé F, Flouri T, McDonald D, Schloss P (2015) Vsearch. Rondon MR, August PR, Bettermann AD et al. (2000) Cloning the soil metagenome: A strategy for accessing the genetic and functional diversity of uncultured microorganisms. Applied and Environmental Microbiology, 66, Sala OE, III FSC, Armesto JJ et al. (2009) Global Biodiversity Scenarios for the Year 2100 Science, 287, Schmidt BR, Kéry M, Ursenbacher S, Hyman OJ, Collins JP (2013) Site occupancy models in the analysis of environmental DNA presence/absence surveys: A case study of an emerging amphibian pathogen. Methods in Ecology and Evolution, 4, Schnell IB, Bohmann K, Gilbert MTP (2015) Tag jumps illuminated - reducing sequence-to-sample misidentifications in metabarcoding studies. Molecular Ecology Resources, 15, Smith DP, Peay KG (2014) Sequence depth, not PCR replication, improves ecological inference from next generation DNA sequencing. PLoS ONE, 9. Taberlet P, Coissac E, Hajibabaei M, Rieseberg LH (2012a) Environmental DNA. Molecular Ecology, 21, Taberlet P, Coissac E, Pompanon F et al. (2007) Power and limitations of the chloroplast trnl (UAA) intron for plant DNA barcoding. Nucleic Acids Research, 35. Taberlet P, Prud Homme SM, Campione E et al. (2012b) Soil sampling and isolation of extracellular DNA from large amount of starting material suitable for metabarcoding studies. Molecular Ecology, 21, Thomsen PF, Kielgast J, Iversen LL et al. (2012a) Detection of a Diverse Marine Fish Fauna Using Environmental DNA from Seawater Samples. PLoS ONE, 7, 1 9. Thomsen PF, Kielgast J, Iversen LL et al. (2012b) Monitoring endangered freshwater biodiversity using environmental DNA. Molecular Ecology, 21, Thomsen PF, Willerslev E (2015) Environmental DNA - An emerging tool in conservation for monitoring past and present biodiversity. Biological Conservation, 183, Tiffin P, Ross-Ibarra J (2014) Swarm v2: highly-scalable and high-resolution amplicon clustering. PeerJ PrePints, Valentini A, Pompanon F, Taberlet P (2008) DNA barcoding for ecologists. Trends in Ecology and Evolution, 24, Venables WN, Ripley BD (2002) Modern Applied Statistics with S. In:. Springer, New York. Virgilio M, Backeljau T, Nevado B, De Meyer M (2010) Comparative performances of DNA barcoding across insect orders. BMC bioinformatics, 11, 206. Wickham H (2009) ggplot2: elegant graphics for data analysis. Springer, New York. Wiens J, Sutter R, Anderson M et al. (2009) Selecting and conserving lands for biodiversity: The role of remote sensing. Remote Sensing of Environment, 113, Wilcox TM, McKelvey KS, Young MK et al. (2013) Robust Detection of Rare Species Using Environmental DNA: The Importance of Primer Specificity. PLoS ONE, 8. 39

40 Willerslev E, Hansen AJ, Binladen J et al. (2003) Diverse Plant and Animal Genetic Records from Holocene and Pleistocene Sediments. Science, 300, Wu T, Ayres E, Li G et al. (2009) Molecular profiling of soil animal diversity in natural ecosystems: Incongruence of molecular and morphological results. Soil Biology and Biochemistry, 41, Yang C, Wang X, Miller JA et al. (2014) Using metabarcoding to ask if easily collected soil and leaf-litter samples can be used as a general biodiversity indicator. Ecological Indicators, 46, Yao H, Song J, Liu C et al. (2010) Use of ITS2 region as the universal DNA barcode for plants and animals. PLoS ONE, 5. Yoccoz NG (2012) The future of environmental DNA in ecology. Molecular Ecology, 21, Yoccoz NG, Bråthen KA, Gielly L et al. (2012) DNA from soil mirrors plant taxonomic and growth form diversity. Molecular Ecology, 21, Yu DW, Ji Y, Emerson BC et al. (2012) Biodiversity soup: Metabarcoding of arthropods for rapid biodiversity assessment and biomonitoring. Methods in Ecology and Evolution, 3, Zhan A, Hulák M, Sylvester F et al. (2013) High sensitivity of 454 pyrosequencing for detection of rare species in aquatic communities. Methods in Ecology and Evolution, 4, Zhou X, Li Y, Liu S et al. (2013) Ultra-deep sequencing enables high-fidelity recovery of biodiversity for bulk arthropod samples without PCR amplification. GigaScience, 2, 4. 40

41 Number of hits or errors Number of hits or errors Supplementary materials Taxonomic assignation and identification rates Average hits at taxonomic levels - DK and full database hits Mahé 3 rep errors Mahé 3 rep hits Mahé 2 rep errors Mahé 2 rep hits OBITools 3 rep errors OBITools 3 rep hits OBITools 2 rep errors OBITools 2 rep Species level Genus level Family level Average hits at taxonomic levels - full database hits Mahé 3 rep errors Mahé 3 rep hits Mahé 2 rep errors Mahé 2 rep hits OBITools 3 rep errors OBITools 3 rep hits OBITools 2 rep errors OBITools 2 rep Species level Genus level Family level Figure S1: Average number of hits and errors by comparison to the Danish restricted and full database (top graph), or the full database only (bottom graph). It is clear that the full database produces more errors and less hits at the species level, while hits at genus and family level remains similar or higher, while error rates are similar or smaller. 41

42 Percentage taxa identified Percentage taxa identified Average ID-rate across nine sample sites - DK & full database Mahé 3 rep Mahé 2 rep OBITools 3 rep OBITools 2 rep -5 species edna / species abgs genera edna / genera abgs families edna / families abgs Average ID-rate across nine sample sites - full database Mahé 3 rep Mahé 2 rep OBITools 3 rep OBITools 2 rep Species edna / species abgs Genera edna / genera abgs Family edna / family abgs Figure S2: The average identification rates across the nine sample sites in the Silkeborg cluster. The percentage of identified species known to occur in a given site was generally higher when the restricted Danish database was included in taxonomic assignment. The trend is also found at the genus level, with the exception of OBITools 2 rep, while the percentage identified families were generally higher when the Danish database was excluded from the taxonomic assignment. 42

43 Species in abg survey Species in abg survey OTU count Species in abg survey OTU count OTU count Rarefaction and species richness Not Obitools db 2 rep: Resampled Read count/1000 Read count/1000 Resampled minimum Not Readcount/1000 OTU count Resampled Resampled minimum OTU count OTU count Figure S3: Examples of rarefaction plots and scatter plots for the correlation between TU count and abg species richness for the OBITools db 2 rep data version. The three blue graphs are rarefaction plots, while the red/purple graphs are correlation plots between abg species richness and OT count. 43

44 Ordinations Figure S4: NMDS ordinations for the abg survey and edna data for the OBITools raw 3 replicates and OBITools db 2 replicates data versions. 44

45 Figure S5: NMDS ordinations for the abg survey and edna data for the OBITools db 3 replicates and Mahé raw 2 replicates data versions. 45

46 Figure S6: NMDS ordinations for the abg survey and edna data for the Mahé raw 3 replicates and Mahé db 2 and 3 replicates data versions. For the Mahé db 3 rep version, outliers have been removed, both extreme and moderate ones. 46

47 Environmental fittings to ordinations Table S1: Envfit results from environmental fittings to the ordination analyses. The axes explaining the environmental parameter the best is shown with r 2 and significance level. The table is continued on the next page. OBITools db 2 rep Environmental parameter Primary axis Secondary axis R 2 and significance level edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** OBITools db 3 rep edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** OBITools raw 3 rep edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** 47

48 Mahé db 2 rep Environmental parameter Primary axis Secondary axis R 2 and significance level edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** Mahé db 3 rep edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** Mahé raw 2 rep edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen *** Abg survey Leaf Nitrogen *** edna Leaf Phosphor *** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** Mahé raw 3 rep edna Median Soil Moisture *** Abg survey Median Soil Moisture *** edna Leaf Nitrogen ** Abg survey Leaf Nitrogen *** edna Leaf Phosphor ** Abg survey Leaf Phosphor *** edna Ellenberg Nitrogen *** Abg survey Ellenberg Nitrogen *** edna Mean Light intensity, day *** Abg survey Mean Light intensity, day *** edna Trees more than 40cm in DBH *** Abg survey Trees more than 40cm in DBH *** 48

49 Species lists Tables S2-S9: Species lists from the sample sites in the Silkeborg cluster. Table S2 - ES069 Knagerne Mahé 3rep: Abg survey Species Genus Family Higher taxa Agrostis capillaris Quercus Oxalidaceae fabids Calamagrostis arundinacea Mahé 2rep: Fagaceae Campylopus introflexus Campylopus pyriformis Species Genus Family Higher taxa Carex pilulifera Fagus sylvatica Quercus Pooideae Oxalidales Carex remota Fagus Oxalidaceae fabids Deschampsia cespitosa Oxalis Fagaceae Deschampsia flexuosa Obitools 3rep: Poaceae Dicranum scoparium Dryopteris carthusiana Species Genus Family Higher taxa Dryopteris dilatata Fagus sylvatica Poa Rosoideae Epilobium angustifolium Quercus Pooideae Fagus sylvatica Obitools 2rep: Fagaceae Poaceae Poaceae Rosaceae Holcus mollis Hypnum jutlandicum Ilex aquifolium Isothecium myosuroides Juncus effusus Larix sp. Luzula pilosa Species Genus Family Higher taxa Maianthemum bifolium Galium boreale Vaccinium Rosoideae asterids Molinia caerulea Salix triandra Typha Pooideae Oxalis acetosella Trifolium repens Quercus Fagaceae Picea abies Carex pallescens Poa Ericaceae Picea sitchensis Calluna vulgaris Galium Typhaceae Polytrichastrum formosum Fagus sylvatica Salix Poaceae Rubus idaeus Phragmites australis Trifolium Salicaceae Stellaria holostea Dactylis glomerata Carex Fabaceae Trientalis europaea Aira praecox Calluna Cyperaceae Aira Dactylis Phragmites Fagus Rubiaceae Rosaceae Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 49

50 Table S3 - ES062 Urfuglebakkevej Mahé 3rep: Abg survey Species Genus Family Higher taxon Calluna vulgaris Calluna vulgaris Calluna Ericaceae Campylopus introflexus Mahé 2rep: Carex pilulifera Species Genus Family Higher taxon Cerastium fontanum Calluna vulgaris Calluna Ericoideae Ceratodon purpureus Pooideae Cytisus scoparius Ericaceae Deschampsia flexuosa Poaceae Dicranum scoparium Obitools 3rep: Empetrum nigrum Species Genus Family Higher taxon Galium saxatile Isothecium myosuroides Calluna Ericaceae Hypnum jutlandicum Calluna vulgaris Salix Salicaceae Luzula multiflora Dicranum scoparium Dicranum Dicranaceae Pilosella officinarum Salix triandra Isothecium Lembophyllaceae Pleurozium schreberi Obitools 2rep: Poa pratensis Species Genus Family Higher taxon Polytrichastrum formosum Salix triandra Vaccinium Antirrhineae Asterales Polytrichum juniperinum Trifolium repens Quercus Rosoideae asterids Potentilla erecta Isothecium myosuroides Populus Pooideae Bryidae Rumex acetosella Elymus repens Prunus Fabaceae Veronica officinalis Calluna vulgaris Genista Salicaceae Vaccinium myrtillus Salix Dicranaceae Fagus sylvatica Dicranum Lembophyllaceae Carex maritima Isothecium Ericaceae Populus alba Trifolium Fagaceae Sherardia arvensis Elymus Poaceae Phragmites australis Calluna Cyperaceae Dicranum scoparium Fagus Rosaceae Hylocomium splendens Carex Sherardia Phragmites Hylocomium Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 50

51 Table S4 - ES065 Tørvefladen Mahé 3rep: Abg survey Species Genus Family Higher taxon Agrostis canina Rhynchospora alba Rhynchospora Cyperaceae Atrichum tenellum Drosera rotundifolia Drosera Droseraceae Atrichum undulatum Mahé 2rep: Bryum tenuisetum Species Genus Family Higher taxon Calluna vulgaris Drosera rotundifolia Drosera Pooideae Campylopus flexuosus Rhynchospora alba Rhynchospora Cyperaceae Campylopus introflexus Carex pallescens Carex Droseraceae Carex demissa Galium boreale Galium Rubiaceae Carex nigra Hydrocotyle verticillata Juncus Juncaceae Carex panicea Hydrocotyle Araliaceae Carex pilulifera Obitools 3rep: Cerastium fontanum Species Genus Family Higher taxon Ceratodon purpureus Calluna vulgaris Calluna Ericaceae Cladopodiella fluitans Atrichum tenellum Sphagnum Sphagnaceae Drosera intermedia Juncus subnodulosus Atrichum Polytrichaceae Drosera rotundifolia Sphagnum subsecundum Juncus Juncaceae Erica tetralix Phragmites australis Phragmites Poaceae Eriophorum angustifolium Drosera rotundifolia Drosera Droseraceae Fossombronia foveolata Rhynchospora alba Rhynchospora Cyperaceae Galium saxatile Obitools 2rep: Hydrocotyle vulgaris Species Genus Family Higher taxon Juncus bulbosus Salix triandra Salix Pooideae Bryopsida Juncus effusus Carex pallescens Carex Ericaceae Luzula multiflora Calluna vulgaris Calluna Sphagnaceae Lycopodiella inundata Atrichum tenellum Sphagnum Polytrichaceae Molinia caerulea Sherardia arvensis Atrichum Juncaceae Narthecium ossifragum Juncus subnodulosus Quercus Poaceae Picea sitchensis Sphagnum subsecundum Sherardia Droseraceae Polytrichastrum longisetum Phragmites australis Juncus Cyperaceae Polytrichum commune Rhynchospora alba Phragmites Salicaceae Populus tremula Drosera rotundifolia Rhynchospora Fagaceae Potentilla erecta Sphagnum fimbriatum Drosera Rubiaceae Rhynchospora alba Rhynchospora fusca Salix repens Salix sp. Sphagnum auriculatum Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 51

52 Table S5 - ES066 Odderholm Mahé 3rep: Abg survey Species Genus Family Higher taxon Agrostis canina Carex pallescens Carex Poeae Pentapetalae Agrostis stolonifera Galium boreale Galium Rosoideae Asterales Ajuga reptans Lysimachia thyrsiflora Juncus Potentilleae Fagales Alnus glutinosa Eriophorum angustifolium Lysimachia Primulaceae asterids Alopecurus geniculatus Centranthus ruber Eriophorum Antirrhineae campanulids Anemone nemorosa Juncus squarrosus Centranthus Betulaceae Apiineae Anthoxanthum odoratum Carex paniculata Trichophorum Cyperaceae Aulacomnium palustre Juncus subnodulosus Rubiaceae Betula pubescens Juncaceae Brachythecium mildeanum Primulaceae Brachythecium rivulare Caprifoliaceae Brachythecium rutabulum Plantaginaceae Briza media Rosaceae Bryum pseudotriquetrum Poaceae Calamagrostis canescens Mahé 2rep: Calliergon cordifolium Species Genus Family Higher taxon Calliergon giganteum Filipendula vulgaris Filipendula Primulaceae Pentapetalae Calliergonella cuspidata Triglochin palustris Triglochin Poeae Asterales Caltha palustris Swartzia arborescens Swartzia Fragariinae asterids Campylium polygamum Carex pallescens Carex Rosoideae campanulids Campylium protensum Filipendula ulmaria Galium Anemoneae lamiids Campylium stellatum Galium boreale Silene Potentilleae Fagales Cardamine pratensis Galium songaricum Juncus Caprifoliaceae eudicotyledons Carex appropinquata Silene flos-cuculi Lysimachia Araliaceae Lamiales Carex canescens Lysimachia thyrsiflora Caltha Plantaginaceae Apiineae Carex demissa Eriophorum angustifolium Eriophorum Antirrhineae Carex echinata Centranthus ruber Centranthus Betulaceae Carex elata Hydrocotyle verticillata Hydrocotyle Asteraceae Carex lepidocarpa Juncus squarrosus Salix Rosaceae Carex nigra Salix triandra Fragaria Juncaginaceae Carex panicea Carex paniculata Trichophorum Cyperaceae Carex paniculata Fragaria viridis Populus Rubiaceae Carex rostrata Juncus subnodulosus Pooideae Carex viridula Caryophyllaceae Cerastium fontanum Juncaceae Cirsium palustre Ranunculaceae Climacium dendroides Poaceae Comarum palustre Fabaceae Crataegus monogyna Salicaceae Cynosurus cristatus Obitools 3rep: Dactylorhiza majalis Species Genus Family Higher taxon Danthonia decumbens Galium boreale Juncus Asteraceae Asterales Deschampsia cespitosa Nymphoides peltata Carex Rosoideae Bryidae Dicranum bonjeanii Carex pallescens Galium Pooideae asterids Dryopteris carthusiana Tragopogon pratensis Nymphoides Primulaceae Epilobium palustre Salix triandra Tragopogon Cyperaceae Equisetum fluviatile Carex maritima Salix Rubiaceae Eriophorum angustifolium Eriophorum angustifolium Eriophorum Juncaceae Eupatorium cannabinum Juncus subnodulosus Lysimachia Salicaceae Festuca rubra Carex paniculata Leptobryum Meesiaceae Filipendula ulmaria Lysimachia thyrsiflora Rubus Rosaceae Fissidens adianthoides Leptobryum pyriforme Centranthus Menyanthaceae Frangula alnus Rubus idaeus Caprifoliaceae Galium palustre Centranthus ruber Poaceae Galium uliginosum Juncus squarrosus Glyceria fluitans Carex otrubae Hamatocaulis vernicosus 52

53 Obitools 2rep: Holcus lanatus Species Genus Family Higher taxon Hydrocotyle vulgaris Galium boreale Juncus Rosoideae Lamiales Juncus articulatus Salix triandra Sphagnum Asteraceae Asterales Juncus conglomeratus Trifolium repens Quercus Apioideae Bryidae Juncus effusus Elymus repens Populus Pooideae asterids Lotus pedunculatus Nymphoides peltata Carex Rosaceae Luzula campestris Carex pallescens Trichophorum Primulaceae Luzula multiflora Tragopogon pratensis Galium Rhinantheae Lychnis flos-cuculi Calluna vulgaris Salix Brachytheciaceae Lycopus europaeus Filipendula ulmaria Trifolium Cyperaceae Lysimachia thyrsiflora Carex flacca Elymus Rubiaceae Lysimachia vulgaris Anthyllis vulneraria Nymphoides Juncaceae Mentha x Carex maritima Tragopogon Salicaceae Menyanthes trifoliata Eriophorum angustifolium Calluna Meesiaceae Molinia caerulea Plagiomnium undulatum Filipendula Menyanthaceae Myosotis laxa Juncus subnodulosus Anthyllis Caprifoliaceae Myrica gale Filipendula vulgaris Eriophorum Sphagnaceae Odontoschisma sphagni Sphagnum subsecundum Plagiomnium Fagaceae Parnassia palustris Juncus squarrosus Lysimachia Fabaceae Pedicularis palustris Carex paniculata Leptobryum Poaceae Peucedanum palustre Lysimachia thyrsiflora Rubus Ericaceae Philonotis fontana Leptobryum pyriforme Hypochaeris Mniaceae Plagiomnium elatum Rubus idaeus Galinsoga Caryophyllaceae Plagiomnium ellipticum Hypochaeris maculata Aira Adoxaceae Plagiomnium undulatum Galinsoga quadriradiata Centranthus Apiaceae Poa pratensis Aira praecox Silene Orobanchaceae Poa trivialis Centranthus ruber Adoxa Prunella vulgaris Silene flos-cuculi Festuca Pseudoscleropodium purum Sphagnum fimbriatum Fragaria Ranunculus acris Carex caryophyllea Ranunculus flammula Carex limosa Ranunculus lingua Adoxa moschatellina Ranunculus repens Festuca pratensis Rhytidiadelphus squarrosus Carex otrubae Rubus idaeus Fragaria viridis Rumex acetosa Salix aurita Salix cinerea Salix pentandra Scorpidium cossonii Sorbus intermedia Sphagnum contortum Sphagnum sp. Sphagnum squarrosum Sphagnum teres Stellaria graminea Stellaria media Taraxacum sp. Thelypteris palustris Tomentypnum nitens Trifolium pratense Trifolium repens Triglochin palustris Valeriana dioica Taxa that were present in the abg survey. Valeriana sambucifolia Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Veronica beccabunga Taxa likely to be a result of database issues discussed later. Veronica scutellata May be cases of "false negatives". Viola palustris Species found to be dominating at the sample site. Warnstorfia exannulata 53

54 Table S6 - ES068 Haarup Sande Mahé 3rep: Abg survey Species Genus Family Higher taxon Abies alba Calluna vulgaris Vaccinium Pinaceae Betula pendula Vaccinium vitis-idaea Calluna Ericaceae Betula pubescens Mahé 2rep: Brachythecium rutabulum Species Genus Family Higher taxon Calluna vulgaris Calluna vulgaris Vaccinium Pinaceae Carex arenaria Vaccinium vitis-idaea Calluna Ericaceae Chiloscyphus coadunatus Betulaceae Crataegus monogyna Obitools 3rep: Crataegus sp. Species Genus Family Higher taxon Deschampsia flexuosa Isothecium myosuroides Vaccinium Pinaceae Dicranum polysetum Calluna vulgaris Isothecium Ericaceae Dicranum scoparium Vaccinium myrtillus Calluna Lembophyllaceae Dryopteris carthusiana Vaccinium vitis-idaea Picea Lophocoleaceae Empetrum nigrum Picea abies Chiloscyphus Fagus sylvatica Chiloscyphus profundus Hylocomium splendens Obitools 2rep: Hypnum cupressiforme Species Genus Family Higher taxon Hypnum jutlandicum Salix triandra Vaccinium Pooideae Asterales Larix kaempferi Trifolium repens Quercus Pinaceae Bryidae Larix sp. Isothecium myosuroides Poa Ericaceae Hypnales Monotropa hypopitys Gymnocarpium dryopteris Carex Lembophyllaceae asterids Picea abies Calluna vulgaris Salix Lophocoleaceae Pinus sylvestris Vaccinium myrtillus Trifolium Fagaceae Pleurozium schreberi Fagus sylvatica Isothecium Poaceae Polytrichum commune Vaccinium vitis-idaea Gymnocarpium Cyperaceae Prunus padus Picea abies Calluna Salicaceae Prunus serotina Chiloscyphus profundus Fagus Fabaceae Pseudoscleropodium purum Hylocomium splendens Picea Cystopteridaceae Quercus robur Chiloscyphus Hylocomiaceae Quercus rubra Hylocomium Sorbus aucuparia Vaccinium myrtillus Vaccinium vitis-idaea Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 54

55 Table S7 - ES064 Mossø Mahé 3rep: Abg survey Species Genus Family Higher taxon Agrostis stolonifera Carex pallescens Juncus Rosoideae eudicotyledons Ajuga reptans Armoracia rusticana Ranunculus Pooideae Pentapetalae Anemone nemorosa Galium boreale Carex Juncaceae Lamiales Angelica sylvestris Armoracia Ranunculaceae Anthoxanthum odoratum Galium Cyperaceae Aulacomnium palustre Brassicaceae Brachythecium rutabulum Rubiaceae Calliergon cordifolium Rosaceae Calliergonella cuspidata Poaceae Caltha palustris Mahé 2rep: Cardamine pratensis Species Genus Family Higher taxon Carex canescens Carex pallescens Rumex Polygonoideae Asterales Carex echinata Poeae Carex nigra Armoracia rusticana Juncus Fragariinae Chloroplast Group 2 (Poeae type) Anthyllis vulneraria Ranunculus Rosoideae Pentapetalae Carex panicea Galium boreale Loteae Pooideae asterids Carex rostrata Fraxinus excelsior Carex Ranunculeae eudicotyledons Cirsium palustre Fragaria viridis Armoracia Betulaceae Lamiales Climacium dendroides Anthyllis Poaceae Gunneridae Comarum palustre Galium Brassicaceae Dactylorhiza majalis Fraxinus Juncaceae Deschampsia cespitosa Fragaria Ranunculaceae Epilobium palustre Cyperaceae Equisetum fluviatile Rubiaceae Eriophorum angustifolium Polygonaceae Festuca rubra Fabaceae Filipendula ulmaria Oleaceae Galium palustre Rosaceae Galium saxatile Obitools 3rep: Galium uliginosum Species Genus Family Higher taxon Glyceria fluitans Nymphoides peltata Juncus Rosoideae Bryidae Holcus lanatus Carex pallescens Ranunculus Pooideae Asterales Iris pseudacorus Fraxinus excelsior Nymphoides Rosaceae asterids Juncus effusus Ranunculus reptans Carex Poaceae Juncus filiformis Ranunculus sceleratus Fraxinus Juncaceae Lotus pedunculatus Armoracia rusticana Armoracia Ranunculaceae Luzula multiflora Rubus idaeus Rubus Cyperaceae Lychnis flos-cuculi Aira praecox Aira Menyanthaceae Lycopus europaeus Elymus repens Elymus Oleaceae Lysimachia thyrsiflora Festuca pratensis Festuca Brassicaceae Lysimachia vulgaris Obitools 2rep: Mentha aquatica Species Genus Family Higher taxon Mentha x Galium boreale Juncus Rosoideae Bryidae Menyanthes trifoliata Salix triandra Poa Asteraceae Asterales Myosotis scorpioides Trifolium repens Quercus Rosaceae asterids Nardus stricta Elymus repens Carex Pooideae Peucedanum palustre Nymphoides peltata Ranunculus Brassicaceae Plagiomnium ellipticum Carex pallescens Galium Poaceae Poa pratensis Fraxinus excelsior Salix Juncaceae Poa trivialis Calluna vulgaris Trifolium Ranunculaceae Ranunculus acris Fagus sylvatica Elymus Cyperaceae Ranunculus flammula Ranunculus reptans Nymphoides Menyanthaceae Ranunculus repens Anthyllis vulneraria Fraxinus Oleaceae Rhytidiadelphus squarrosus Carex maritima Calluna Fagaceae Rumex acetosa Blechnum spicant Fagus Rubiaceae Scutellaria galericulata 55

56 Juncus squarrosus Anthyllis Salicaceae Stellaria palustris Phragmites australis Blechnum Fabaceae Trifolium repens Ranunculus sceleratus Phragmites Ericaceae Veronica scutellata Dactylis glomerata Dactylis Blechnaceae Viola palustris Armoracia rusticana Armoracia Cornaceae Rubus idaeus Rubus Crataegus monogyna Crataegus Aira praecox Aira Cornus suecica Cornus Poa compressa Festuca pratensis Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 56

57 Table S8 - ES070 Gjessøvej Mahé 3rep: Abg survey Species Genus Family Higher taxon Acer pseudoplatanus Salix triandra Quercus Primulaceae Asterales Agrostis stolonifera Lysimachia thyrsiflora Ranunculus Potentilleae Malpighiales Anemone nemorosa Carex Solanoideae eudicotyledons Aneura maxima Populus Betulaceae Fagales Angelica sylvestris Salix Rosaceae Betula pendula Lysimachia Solanaceae Betula pubescens Fagaceae Brachythecium rivulare Ranunculaceae Brachythecium rutabulum Cyperaceae Calliergonella cuspidata Salicaceae Cardamine flexuosa Mahé 2rep: Cardamine pratensis Species Genus Family Higher taxon Carex nigra Carex pallescens Ranunculus Primulaceae Asterales Chiloscyphus coadunatus Salix triandra Quercus Poaceae Fagales Chiloscyphus pallescens Lysimachia thyrsiflora Carex Solanaceae Malpighiales Chiloscyphus profundus Fagus Rosoideae asterids Cirsium palustre Populus Potentilleae eudicotyledons Comarum palustre Salix Pooideae Pentapetalae Crataegus sp. Lysimachia Solanoideae Deschampsia cespitosa Betulaceae Dicranum scoparium Rosaceae Dryopteris carthusiana Fagaceae Dryopteris dilatata Ranunculaceae Epilobium palustre Cyperaceae Equisetum fluviatile Salicaceae Equisetum palustre Obitools 3rep: Crataegus sp. Species Genus Family Higher taxon Fagus sylvatica Salix triandra Juncus Rosoideae Asterales Festuca rubra Gymnocarpium dryopteris Sphagnum Brachytheciaceae Bryidae Frangula alnus Fagus sylvatica Populus Primulaceae Frullania dilatata Anthyllis vulneraria Quercus Juncaceae Galium aparine Plagiomnium undulatum Salix Sphagnaceae Galium palustre Chiloscyphus profundus Gymnocarpium Salicaceae Galium uliginosum Sphagnum subsecundum Fagus Fagaceae Holcus lanatus Lysimachia thyrsiflora Anthyllis Cystopteridaceae Hylocomium splendens Chiloscyphus polyanthos Plagiomnium Fabaceae Hypnum cupressiforme Mnium hornum Chiloscyphus Mniaceae Juncus conglomeratus Lysimachia Lophocoleaceae Juncus effusus Mnium Rosaceae Kindbergia praelonga Obitools 2rep: Lonicera periclymenum Species Genus Family Higher taxon Lotus pedunculatus Galium boreale Juncus Rosoideae asterids Lysimachia vulgaris Salix triandra Sphagnum Pooideae Asterales Metzgeria fruticulosa Gymnocarpium dryopteris Quercus Brachytheciaceae Bryidae Mnium hornum Nymphoides peltata Populus Primulaceae Molinia caerulea Carex pallescens Carex Juncaceae Orthotrichum affine Calluna vulgaris Meesia Sphagnaceae Orthotrichum pulchellum Fagus sylvatica Galium Salicaceae Pellia endiviifolia Rhynchostegium riparioides Salix Fagaceae Peucedanum palustre Anthyllis vulneraria Phragmites Cystopteridaceae Plagiomnium elatum Plagiomnium undulatum Gymnocarpium Fabaceae Plagiomnium ellipticum Chiloscyphus profundus Nymphoides Mniaceae Plagiomnium undulatum Sphagnum subsecundum Juglans Lophocoleaceae Plagiothecium denticulatum Juglans regia Calluna Rosaceae Plagiothecium undulatum Phragmites australis Adoxa Cyperaceae Pleurozium schreberi Mnium hornum Fagus Meesiaceae Poa trivialis 57

58 Lysimachia thyrsiflora Rhynchostegium Rubiaceae Polytrichastrum formosum Chiloscyphus polyanthos Anthyllis Poaceae Polytrichum commune Adoxa moschatellina Mnium Menyanthaceae Pseudoscleropodium purum Sphagnum fimbriatum Lysimachia Juglandaceae Ranunculus repens Plagiomnium Ericaceae Rhizomnium punctatum Chiloscyphus Adoxaceae Rhytidiadelphus squarrosus Rhytidiadelphus triquetrus Rubus sect. Salix cinerea Sambucus nigra Sanionia uncinata Solanum dulcamara Sorbus aucuparia Sphagnum fimbriatum Sphagnum palustre Sphagnum russowii Sphagnum squarrosum Stellaria holostea Taraxacum sp. Ulota bruchii Urtica dioica Valeriana sambucifolia Veronica scutellata Viola palustris Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 58

59 Table S9 - ES067 Højkol-Rye Sønderskov Mahé 3rep: Abg survey Species Genus Family Higher taxon Betula pubescens Pteridium Betulaceae Brachythecium rutabulum Dennstaedtiaceae Calluna vulgaris Mahé 2rep: Campylopus introflexus Species Genus Family Higher taxon Campylopus pyriformis Fagus sylvatica Quercus Cupressaceae Cephaloziella divaricata Calluna vulgaris Pteridium Ericaceae Ceratodon purpureus Vaccinium vitis-idaea Fagus Pooideae Cytisus scoparius Calluna Betulaceae Deschampsia flexuosa Vaccinium Dennstaedtiaceae Dicranum polysetum Poaceae Dicranum scoparium Fagaceae Epilobium angustifolium Obitools 3rep: Fagus sylvatica Species Genus Family Higher taxon Frangula alnus Isothecium myosuroides Vaccinium Grimmiaceae Bryidae Galium saxatile Calluna vulgaris Quercus Ericaceae Hypnales Holcus mollis Vaccinium myrtillus Isothecium Fagaceae Hylocomium splendens Fagus sylvatica Calluna Lembophyllaceae Hypnum cupressiforme Dicranum scoparium Fagus Dicranaceae Hypnum jutlandicum Hylocomium splendens Dicranum Hylocomiaceae Juniperus communis Chamaecyparis lawsoniana Hylocomium Cupressaceae Lonicera periclymenum Hamatocaulis vernicosus Chamaecyparis Calliergonaceae Picea abies Vaccinium vitis-idaea Hamatocaulis Pinus sylvestris Obitools 2rep: Pleurozium schreberi Species Genus Family Higher taxon Polytrichastrum formosum Salix triandra Vaccinium Pooideae Asterales Prunus serotina Isothecium myosuroides Sphagnum Brachytheciaceae Bryidae Pseudoscleropodium purum Gymnocarpium dryopteris Quercus Grimmiaceae Hypnales Pteridium aquilinum Elymus repens Carex Ericaceae asterids Quercus robur Carex pallescens Meesia Fagaceae Ribes uva-crispa Calluna vulgaris Salix Lembophyllaceae Rubus idaeus Vaccinium myrtillus Isothecium Dicranaceae Rumex acetosella Fagus sylvatica Gymnocarpium Hylocomiaceae Sorbus aucuparia Vaccinium vitis-idaea Elymus Cupressaceae Sorbus intermedia Chamerion angustifolium Calluna Calliergonaceae Trientalis europaea Phragmites australis Fagus Sphagnaceae Ulota bruchii Dicranum scoparium Chamerion Cyperaceae Vaccinium myrtillus Hylocomium splendens Phragmites Meesiaceae Chamaecyparis lawsoniana Dicranum Salicaceae Hamatocaulis vernicosus Hylocomium Cystopteridaceae Chamaecyparis Poaceae Hamatocaulis Onagraceae Taxa that were present in the abg survey. Taxa ranked as tribes etc. below the family level, but above genus. These were only part of the analysis through the matching family name. Taxa likely to be a result of database issues discussed later. May be cases of "false negatives". Species found to be dominating at the sample site. 59

60 Species richness in abg survey OTU count OTU count Correlation plots between OTU count and environmental parameters Table S10: The edna dataset versions showing significant correlations with median soil moisture. Median Soil Moisture T df 95 CI p-value cor OBITools raw 2 OTU count not rep OTU count minimum Abg species richness OBITools db 3 rep OTU count not OTU count minimum Abg species richness No significant correlations were found for OBITools raw 3 rep, Mahé raw 2 rep, Mahé raw 3 rep, Mahé db 2 rep and Mahé db 3 rep. Not OBITools db 3 rep Resampled minimum Mean Soil moisture Mean Soil moisture Abg survey Figure S6: Scatter plots, here exemplified by the OBITools db 3 replicates data version. Both the OTU count minimum and abgs yield significant correlations. Correlation values can be seen in table S10. Mean Soil moisture 60

61 Table S11: The edna dataset versions showing significant correlations with soil ph. Mahé db 2 rep Mahé db 3 rep Mahé raw 3 rep OBITools db 3 rep OBITools raw 2 rep OBITools raw 3 rep Soil ph T df 95 CI p-value cor OTU count not e OTU count minimum Abg species richness e OTU count not e OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e

62 Species richness in abg survey OTU count OTU count Not OBITools raw 2 rep Resampled minimum Soil ph Soil ph Abg survey Figure S7: Scatter plots of richness estimates and soil ph, here exemplified by the OBITools raw 2 rep data version. All correlations are significant. Correlation values can be seen in table S11 Soil ph 62

63 Table S12: The edna dataset versions showing significant correlations with mean surface temperature. Mahé db 2 rep Mahé db 3 rep Mahé raw 3 rep OBITools db 3 rep OBITools raw 2 rep OBITools db 2 rep Mean surface temperature T df 95 CI p-value cor OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness

64 OTU count OTU count Species richness abgs OBITools db 3 rep Abg survey Figure S8: Scatter plots of richness estimates and surface temperature, here exemplified by the OBITools db 3 rep data version. Correlation values can be seen in table S12. Surface temperature Not Resampled minimum Surface temperature Surface temperature 64

65 Table S13: The edna dataset versions showing significant correlations with organic matter ph. Mahé db 2 rep Mahé db 3 rep Mahé raw 2 rep Mahé raw 3 rep OBITools db 2 rep OBITools db 3 rep OBITools raw 2 rep OBITools db 2 rep Organic matter ph T df 95 CI p-value cor OTU count not e OTU count minimum Abg species richness e OTU count not e OTU count minimum Abg species richness OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e OTU count not OTU count minimum Abg species richness e

66 OTU count OTU count Mahé raw 3 rep Abg survey Figure S9: Scatter plots of richness estimates and Organic matter ph, here exemplified by the Mahé raw 3 rep data version. All correlations are significant. Correlation values can be seen in table S13. Organic matter ph Not Resampled minimum Organic matter ph Organic matter ph 66

67 Residuals and environmental parameters Table S14: An overview of which pipeline versions and re-sample types correlated significantly with a given environmental parameter. Significant correlation results can be seen in table S15. Parameter Pipeline Sampling type: Organic Matter Mean surface version Soil ph Soil moisture ph Temp. Obitools db 3 Not re-sampled x Re-sampled Re-sampled minimum Obitools db 2 Not re-sampled x Re-sampled x Re-sampled minimum x Obitools raw 3 Not re-sampled x Re-sampled x Re-sampled minimum x Obitools raw 2 Not re-sampled x Re-sampled Re-sampled minimum Mahé db 3 Not re-sampled x x x x Re-sampled x x Re-sampled minimum x Mahé db 2 Not re-sampled x x x Re-sampled x Re-sampled minimum Mahé raw 3 Not re-sampled x Re-sampled Re-sampled minimum Mahé raw 2 Not re-sampled Re-sampled Re-sampled minimum 67

68 Table S15: Results from correlations between environmental parameters and the residuals from a linear model between OUT count and above ground species richness. Pipeline version Sampling type: Parameter t df p-value Obitools db 3 Obitools db 2 Not re-sampled Not re-sampled Re-sampled Re-sampled minimum Mean surface Temp. Organic Matter ph Organic Matter ph 95%-confidence interval Correlationcoefficient Soil moisture Obitools raw 3 Not re-sampled Mean surface Temp Re-sampled Mean surface Temp Re-sampled Mean surface minimum Temp Obitools raw 2 Not re-sampled Mean surface Temp Mahé db 3 Not re-sampled Soil ph Re-sampled Soil ph Re-sampled minimum Soil ph Not re-sampled Organic Matter ph Re-sampled Organic Matter ph Not re-sampled Soil moisture Not re-sampled Mean surface Temp Mahé db 2 Not re-sampled Soil ph Re-sampled Organic Matter ph Not re-sampled Soil moisture Not re-sampled Mean surface Temp Mahé raw 3 Not re-sampled Soil ph

69 Scripts Script for reference database construction #To make the full trnl database: # THIS IS IN BASH/LINUX + OBITOOLS #Based on tutorial here: (uses EMBL data, I cannot get that to work, so uses GenBank instead where the plant sequences are called gbplnxxxxxxseq.gz and are placed in ftp://ftp.ncbi..nlm.nih.gov/genbank/) #Download plant sequences from Genbank (NCBI) mkdir NCBI cd NCBI nice -n 19 wget ftp://ftp.ncbi.nlm.nih.gov/genbank/gbpln*seq.gz cd.. #Download the taxonomy from NCBI mkdir TAXO cd TAXO nice -n 19 wget ftp://ftp.ncbi.nih.gov/pub/taxonomy/taxdump.tar.gz tar -zxvf taxdump.tar.gz cd.. #Format the data mkdir ncbi_database cd ncbi_database nice -n 19 obiconvert --genbank -t../taxo --ecopcrdb-output=ncbi_last../ncbi/*.seq.gz #Use ecopcr to simulate an in silico PCR nice -n 19 ecopcr -d ncbi_last -e 3 -l 8 -L 160 GGGCAATCCTGAGCCAA CCATTGAGTCTCTGCACCTATC > newtrnl01.ecopcr #filter sequences so that they have a good taxonomic description at the species, genus, and family levels (obigrep command below). obigrep -d ncbi_last --require-rank=species --require-rank=genus --require-rank=family newtrnl01.ecopcr > newtrnldb_clean.fasta #remove redundant sequences (obiuniq command below). obiuniq -d ncbi_last newtrnldb_clean.fasta > newtrnldb_clean_uniq.fasta #ensure that the dereplicated sequences have a taxid at the family level (obigrep command below). obigrep -d ncbi_last --require-rank=family newtrnldb_clean_uniq.fasta > newtrnldb_clean_uniq_clean.fasta #ensure that sequences each have a unique identification (obiannotate command below) obiannotate --uniq-id newtrnldb_clean_uniq_clean.fasta > newtrnldb_v01.fasta #To construct the danish database from full-database: #IN R: #Import the "names.dmp" file from the downloaded taxonomy. The first column is the taxid, while the third column is the scientific name. 69

70 names_dump <- read.delim("path/names.dmp", header = TRUE,stringsAsFactors=FALSE) #import vascular plant list. The first column is the scientific name. karplanter_list <- read.csv("path/allearter_karplanter.csv", sep = ";") #import bryophyte list. The first column is the scientific name. mosses_list <- read.csv("path/allearter_mosser.csv", sep = ";") #Merge the two dataframes with species lists: species_list <- rbind(karplanter_list,mosses_list) #Export scientific name column to vector: species_list <- species_list[,1] # a character is made, and a loop finds the scientific name in the "names.dmp" file, and if there is a match, it writes the taxid to the character "species_list_taxid". If there is no match, it writes "NA". species_list_taxid = character(0) for (species in species_list) { if (species %in% names_dump[,3]){ species_list_taxid <- append(species_list_taxid, names_dump[which(names_dump[,3]==species),1])} else species_list_taxid <- append(species_list_taxid, "NA")} # "NA"'s are removed: and the resulting list of taxids are just barely 2000 entries long, so a reduction of approximately 1300 species or subspecies/variants. species_list_taxid_nona <- species_list_taxid[-which(species_list_taxid=="na")] #The list is exported out=paste(path/taxidlist_danishspecies_nona.csv',sep='') write.csv(species_list_taxid_nona,out) #THIS IS IN BASH/LINUX #The taxid list has been uploaded to the server and is about to enter a loop, where every line is being searched for as taxid in the full database, and matches are stores in a new file. For this to happen I change the appearence of the taxids a little bit: aagaard@zurhausen:~/data/biowide/trnl/db$ awk '{print "^"$0}' taxids_danishsp_nona > taxids_danishsp_nona_beg aagaard@zurhausen:~/data/biowide/trnl/db$ awk '{print $0"$"}' taxids_danishsp_nona_beg > taxids_danishsp_nona_beg_end # This is to add ^ to the beginning of each taxid, and a $ to the end of each taxid. They signify the beginning and end of an entry, respectively. This makes sure it doesn't match random part of numbers, but only matches the whole taxid-number. #Now enter loop, using obigrep, and a command to treat others on the server nicely. aagaard@zurhausen:~/data/biowide/trnl/db$ while read line; do nice -n19 obigrep -a taxid:$line newtrnldb_v01.fasta >> Database_danish_out.fasta; done < taxids_danishsp_nona_beg_end #The sequence-count of the full database-file aagaard@zurhausen:~/data/biowide/trnl/db$ obicount newtrnldb_v01.fasta newtrnldb_v01.fasta 69.3 % ################################## ] remain : 00:00: #The sequence-count of the Danish database-file 70

71 obicount Database_danish_out.fasta Database_danish_out.fasta 94.4 % ###############################################\ ] remain : 00:00: #The output-file is the new danish database. Script for raw data processing the OBITools pipeline An example of the pipeline from raw data to sequences assigned to the given samples #!/bin/bash #trnl sequences workflow script #use nice -n19 on commands #Files from run on number also called sequencing run 1 (seq1) #ESW-UNRF-trnL-R1A_S1_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R1A_S1_L001_R2_001.fastq.gz #ESW-UNRF-trnL-R1B_S2_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R1B_S2_L001_R2_001.fastq.gz #ESW-UNRF-trnL-R2A_S3_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R2A_S3_L001_R2_001.fastq.gz #ESW-UNRF-trnL-R2B_S4_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R2B_S4_L001_R2_001.fastq.gz #ESW-UNRF-trnL-R3A_S5_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R3A_S5_L001_R2_001.fastq.gz #ESW-UNRF-trnL-R3B_S6_L001_R1_001.fastq.gz #ESW-UNRF-trnL-R3B_S6_L001_R2_001.fastq.gz #These are the files from sequencing: /emc/miseq1/151218_fastq/esw* #They can be found in dir: /home/aagaard/data/biowide/trnl/analyses/seq1/rawdat #HERE STARTS SCRIPT #RUN like this: trnl_seq_obiworkflow_script.sh $1 $2 $3 $4 #RUN THIS BEFORE START: Open obitools 1.2.2: # /usr/local/src/obitools/obitools #$1 Forward strand path (in this case /home/aagaard/data/biowide/trnl/analyses/seq1/rawdat/esw-unrf-trnl-r1a_s1_l001_r1_001.fastq.gz) #$2 Reverse strand path (itc: /home/aagaard/data/biowide/trnl/analyses/seq1/rawdat/esw- UNRF-trnL-R1A_S1_L001_R2_001.fastq.gz) #$3 Basename (in this case trnl_seq1_r1a) #$4 NGSfilter path #Files merged/paired, minimum match of 40%: More: # nice -n19 illuminapairedend --score-min=40 -r $1 $2 > $3_paired.fastq #Remove unalligned/paired/matched sequences: more about obigrep: # nice -n19 obigrep -p 'mode!="joined"' $3_paired.fastq > $3_paired_alligned.fastq 71

72 #Assign the sequences to samples via ngsfilter-file: -t specifies the list with the tags inside, ie. the ngsfilterfile. -u specifies which file to store the unidentified seq in. See more: nice -n19 ngsfilter -t $4 -u $3_paired_alligned_unid.fastq $3_paired_alligned.fastq > $3_paired_alligned_assigned.fastq #get rid of comments other than samplename #add identifier with count #add new tag to sequences (sequence replicate): "seq_rep:*" nice -n19 obiannotate -k sample $3_paired_alligned_assigned.fastq nice -n19 obiannotate --set-identifier '"trnlr1a_%07d" % counter' nice -n19 obiannotate -S 'seq_rep:r2' > $3_paired_alligned_assigned_r2_anot.fastq An example of how the pipeline proceses to database assignation and table creation #!/bin/bash #Script for combining all sequence reads from all libraries in one sequencing run and add seqr1 (sequencing replicate 1) og seqr2 #It also dereplicates and cleans the file leading it all the way to a table for opening in excell. It annotates sequences to taxon first in danish db, then the ones with no hits are compared to the full db. #Run like this: trnl_seq_obiworkflow_combine_to_tag_forseq2.sh $1 $2 $3 $4 $5 #run in analyses folder #$1 Basename for outputfile: itc: something that ends like this, since these were the preceeding steps (trnl_seq2_total_paired_alligned_assigned_anot) #$2 databse taxonomy path (/home/aagaard/data/biowide/taxonomy/ncbi_last/ncbi_last) #$3 database path to danish database (/home/aagaard/data/biowide/trnl/db/newtrnl_db_danish_clean_uniq_clean_anot.fasta) #$4 database path to full database (/home/aagaard/data/biowide/trnl/db/newtrnldb_v01.fasta) #$5 Final name for file containing annotated sequences (trnl_seq2_tagged) #$6 Input file (...merged_uniq.fasta) nice -n19 cp -a $6 $1_merged_uniq_WITHrare.fasta #filter out the sequences below 10 bp nice -n19 obigrep -l 10 $1_merged_uniq_WITHrare.fasta > $1_merged_uniq_WITHrare_l10.fasta #remove stepwise pcr errors, single linkage clustering nice -n19 obiclean -r s merged_sample -H $1_merged_uniq_WITHrare_l10.fasta > $1_merged_uniq_WITHrare_l10_clean25.fasta #Taxonomic assignment in danish database nice -n19 ecotag -d $2 -R $3 -m 0.9 $1_merged_uniq_WITHrare_l10_clean25.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag.fasta #grep all sequences where there is no tag at family level or more, ie. no species, genus or family tag: nice -n19 obigrep -a 'rank:no rank' -a 'rank:tribe' -a 'rank:order' $1_merged_uniq_WITHrare_l10_clean25_DANISHtag.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none.fasta #Grep all sequences with family id number, ie. all sequences are tagget to family level or more: this should be the ones with sufficient hit in danish database. 72

73 nice -n19 obigrep -a 'family:^[0-9]+$' $1_merged_uniq_WITHrare_l10_clean25_DANISHtag.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_fam.fasta #give new attribute: databse: danish nice -n19 obiannotate -S 'database:danish' $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_fam.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_fam_anot.fasta #Remove attributes from nontagged seqs: nice -n19 obiannotate -k merged_sample -k obiclean_head -k obiclean_cluster -k count -k obiclean_internalcount -k obiclean_status -k obiclean_samplecount -k obiclean_headcount -k seq_rep -k obiclean_singletoncount $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot.fasta # try tagging no_rank species in full database nice -n19 ecotag -d $2 -R $4 -m 0.9 $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot_FULLtag.fasta #give new attribute: databse: full nice -n19 obiannotate -S 'database:full' $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot_FULLtag.fasta > $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot_FULLtag_anot.fasta # Merge files with tag(annotated) sequences: both danish and full nice -n19 cat $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_fam_anot.fasta >> $5_merged.fasta nice -n19 cat $1_merged_uniq_WITHrare_l10_clean25_DANISHtag_none_anot_FULLtag_anot.fasta >> $5_merged.fasta #Make af OTU table nice -n19 obitab -o $5_merged.fasta > $5_merged_table.txt #check length distribution #obistat -c 'len(sequence)' trnl_total_assigned.clean.uniq.withrare.obiclean.fasta 73

74 Lab work, field work and other endeavors not included in the thesis manuscript As not every doing during the course of a masters project can be included in the main part of the thesis manuscript, here comes a brief overview of the endeavors and thoughts that were left out. Initial lab work When starting out the work on the master thesis, the soil samples from the Biowide project had already been collected and stored in the freezer basement at the Centre for GeoGenetics in Copenhagen. The next step in the edna workflow was extraction but which extraction method should be used? We tested two different extraction kits on 8 different soils, the MoBio PowerMax Soil and Nucleospin Soil kit, with several variations of the Nucleospin kit. A kit was chosen based on PCR amplification of the extracted DNA band strength on agarose gels, and the MoBio PowerSoil won the competition as best extraction kit across the sampled soil types. It took about one and a half month to get acquainted with the laboratory procedures and rules, and extracting and deciding which extraction procedure to use on the 130 Biowide samples. Unfortunately, a lot of waiting followed the primary extractions, since MoBio had discovered a contamination in the PowerMax Soil kit, leading to considerable delays before the kits were delivered but hopefully with no contaminants! Extraction of the 130 soils followed quite a substantial task in a relatively small laboratory and with a large kit using 50mL tubes. Up to 28 samples could be processed within an approximately 12 hour day. In this study, the plant was to amplify two plant sequences that was also part of the Biowide study: the nuclear Internal Transcribed Spacer-2 and the chloroplast trnl p6 loop. Middle: an example of a keying character. Left and right: two of the Carex species sampled, Carex pulicaris and Carex arenaria. Field work and subsequent lab work At this point, the flowering season had started, and I had decided to engage in collecting fresh specimens for a reference database for either the ITS-2 or trnl markers. Furthermore, I also thought about looking at intraspecific variation in the ITS-2 sequences within the Carex genus, sequenced on the Illumina platform together with some of the markers used in Biowide (eg. trnl or ITS-2). So I found a way to collect leaf samples from different Carex and some trees not present in the ncbi database or not present with sequences collected from within Europe. In total I collected 64 silica dried samples, accompanied by 74

75 photographs documenting species ID following the identification key for the Danish Cyperaceae species (Schou 2006), and collected and dried the plants for reference. The collected specimen samples were extracted in the fall 2015, using the Quiagen plant mini kit or the Quiagen blood and tissue kit, and subsequently PCR amplified with tagged primers amplifying the trnl p-loop sequence and the ITS-2 sequence, as these were the ones used on the 130 Biowide soil samples. The amplified DNA was pooled, taking the relative abundance into account, and built into a sequencing library. While I was collecting specimen samples and extracting them, the amplification and sequencing preparation of the trnl p6-loop and plant ITS-2 sequences had been done by a research assistant. When my specimen samples were ready for sequencing, the resulting library was pooled with the Biowide plant ITS-2 libraries, albeit in lower quantities, and submitted for sequencing on the Illumina Miseq sequencer. How to process vast amounts of data and learning three programming languages While waiting for the sequencing results I decided to learn how to treat Next Generation Sequencing raw data. This included learning to use commandlines and learn a few concepts of programming. Many headaches and a few tutorials later, I learned how to manage my own PC using command lines (at least a few simple tasks). It felt a little like magic when typing mkdir in the commandline shell, while looking at it appear in the visual file explorer. However, soon enough I learned that the servers available for raw data processing was Linux based. Not all commands are similar between Linux and Windows, and more internet searches and tutorials followed, and quickly enough I was relatively familiar with the Linux based servers. Unfortunately, the sequencing results from the ITS-2 sequences and my specimen samples did not finish successfully until the middle of February. As a consequence, I decided to begin the raw data processing of the trnl sequences that finished by the end of January, despite the fact that I would rather have looked at the ITS-2 sequences probably due to the known intraspecific variation in the sequences and higher taxonomic accuracy. However, in hindsight, I had enough challenges by just looking at the trnl p6 loop sequences. During the course of raw data processing, I learned how to script while getting a better understanding of PCR and sequencing errors, and sequence specific decisions to be made in such a data processing pipeline. When finally having produced an OTU-table I first tried using Microsoft Excel for further data processing. When realizing that some processes would be very time consuming and click-demanding in excel, I learned to make scripts in Excel s commandline-based Visual Basics. This repeatedly either crashed or froze my computer, and the realization that I had to learn yet another language for data processing struck me. However, now, when I am acquainted with R and R scripts, the idea of using Excel seems silly. Time constraints Another lesson learned from this project is that most of the time you overestimate how many thing you can do within a specific time frame. I did not manage to squeeze in the analysis of the specimen sequences and adding them to the database used. Nor did I have time to look at intraspecific variation within the ITS-2 sequences from the collected specimens, and no time was left for comparing the performance of ITS-2 and trnl p6 loop on the Biowide samples. However, I did try the whole environmental DNA workflow, from accompanying Tobias Frøslev when collecting soil samples in a few sites, over extraction, PCR and library built, to raw data processing and data analysis. In the process I learned three programming languages, sampled my own reference specimens and last but not least, learned a lot about soil environmental DNA. 75

76 Topleft: a step in the extraction of specimen samples. Top middle and right: The Polymerase Chain Reaction machine and setup. Bottom left: The Agarose gel electrophoresis. Bottom right: Reading the gel electrophoresis, the bands visible are DNA and size ladders. References: Schou JC (2006) Danmarks halvgræsser, 2. rev. udg., BFNs forlag. 76